chore: bump version to 4.11.0

- Implemented a full test suite for ContextManager covering initialization, message processing, token-based compression, and error handling.
- Added tests for ContextTruncator focusing on message fixing, truncation by turns, dropping oldest turns, and halving.
- Ensured that both test suites validate edge cases and maintain expected behavior with various message types, including system and tool messages.

.github/copilot-instructions.md

@@ -15,7 +15,6 @@ Always reference these instructions first and fallback to search or bash command
 ### Running the Application
 - Run main application: `uv run main.py` -- starts in ~3 seconds
 - Application creates WebUI on http://localhost:6185 (default credentials: `astrbot`/`astrbot`)
-- Application loads plugins automatically from `packages/` and `data/plugins/` directories
 
 ### Dashboard Build (Vue.js/Node.js)
 - **Prerequisites**: Node.js 20+ and npm 10+ required
@@ -35,7 +34,7 @@ Always reference these instructions first and fallback to search or bash command
 - **ALWAYS** run `uv run ruff check .` and `uv run ruff format .` before committing changes
 
 ### Plugin Development
-- Plugins load from `packages/` (built-in) and `data/plugins/` (user-installed)
+- Plugins load from `astrbot/builtin_stars/` (built-in) and `data/plugins/` (user-installed)
 - Plugin system supports function tools and message handlers
 - Key plugins: python_interpreter, web_searcher, astrbot, reminder, session_controller
 

.github/workflows/stale.yml

@@ -1,27 +1,64 @@
-# This workflow warns and then closes issues and PRs that have had no activity for a specified amount of time.
+# 本工作流用于标记并关闭长期不活跃的 Issue。
+# 目前仅针对带 `bug` 标签的 Issue 生效，不会处理 PR。
 #
-# You can adjust the behavior by modifying this file.
-# For more information, see:
-# https://github.com/actions/stale
-name: Mark stale issues and pull requests
+# 文档: https://github.com/actions/stale
+name: Mark stale bug issues
 
 on:
   schedule:
-  - cron: '21 23 * * *'
+    # 每天 UTC 08:30 执行 (北京时间 16:30)
+    - cron: '30 8 * * *'
+  workflow_dispatch:
+    inputs:
+      dry-run:
+        description: '仅预览, 不实际执行 (Dry run mode)'
+        required: false
+        default: true
+        type: boolean
 
 jobs:
   stale:
-
     runs-on: ubuntu-latest
     permissions:
       issues: write
-      pull-requests: write
 
     steps:
-    - uses: actions/stale@v10
-      with:
-        repo-token: ${{ secrets.GITHUB_TOKEN }}
-        stale-issue-message: 'Stale issue message'
-        stale-pr-message: 'Stale pull request message'
-        stale-issue-label: 'no-issue-activity'
-        stale-pr-label: 'no-pr-activity'
+      - uses: actions/stale@v10
+        with:
+          repo-token: ${{ secrets.GITHUB_TOKEN }}
+          operations-per-run: 200
+
+          # 只处理带 bug 标签的 Issue
+          any-of-labels: 'bug'
+
+          # 不处理 PR
+          days-before-pr-stale: -1
+          days-before-pr-close: -1
+
+          # 不活跃判定与关闭策略: 先标记 stale, 再延迟关闭
+          days-before-issue-stale: 60
+          days-before-issue-close: 30
+
+          stale-issue-label: 'stale'
+          stale-issue-message: |
+            This issue has been automatically marked as **stale** because it has not had any activity.
+            It will be closed in a certain period of time if no further activity occurs.
+            If this issue is still relevant, please leave a comment.
+
+            ---
+
+            该 Issue 已较长时间无活动, 已被标记为 `stale`。
+            如无后续活动, 将在一段时间后自动关闭。
+            如仍需跟进, 请回复评论。
+          close-issue-message: |
+            This issue has been automatically closed due to inactivity.
+            If the problem still exists, feel free to reopen or create a new issue with updated information.
+
+            ---
+
+            该 Issue 因长期无活动已自动关闭。
+            如问题仍存在, 欢迎补充复现信息并重新打开或新建 Issue。
+
+          remove-stale-when-updated: true
+
+          debug-only: ${{ github.event_name == 'workflow_dispatch' && inputs.dry-run }}

.gitignore

@@ -24,9 +24,9 @@ configs/session
 configs/config.yaml
 cmd_config.json
 
-# Plugins and packages
+# Plugins
 addons/plugins
-packages/python_interpreter/workplace
+astrbot/builtin_stars/python_interpreter/workplace
 tests/astrbot_plugin_openai
 
 # Dashboard

README.md

@@ -132,6 +132,7 @@ uv run main.py
 
 **社区维护**
 
+- [Matrix](https://github.com/stevessr/astrbot_plugin_matrix_adapter)
 - [KOOK](https://github.com/wuyan1003/astrbot_plugin_kook_adapter)
 - [VoceChat](https://github.com/HikariFroya/astrbot_plugin_vocechat)
 - [Bilibili 私信](https://github.com/Hina-Chat/astrbot_plugin_bilibili_adapter)
@@ -208,6 +209,7 @@ pre-commit install
 - 5 群：822130018
 - 6 群：753075035
 - 7 群：743746109
+- 8 群：1030353265
 - 开发者群：975206796
 
 ### Telegram 群组

README_en.md

@@ -134,6 +134,7 @@ Or refer to the official documentation: [Deploy AstrBot from Source](https://ast
 
 **Community Maintained**
 
+- [Matrix](https://github.com/stevessr/astrbot_plugin_matrix_adapter)
 - [KOOK](https://github.com/wuyan1003/astrbot_plugin_kook_adapter)
 - [VoceChat](https://github.com/HikariFroya/astrbot_plugin_vocechat)
 - [Bilibili Direct Messages](https://github.com/Hina-Chat/astrbot_plugin_bilibili_adapter)

README_fr.md

@@ -134,6 +134,7 @@ Ou consultez la documentation officielle : [Déployer AstrBot depuis les sources
 
 **Maintenues par la communauté**
 
+- [Matrix](https://github.com/stevessr/astrbot_plugin_matrix_adapter)
 - [KOOK](https://github.com/wuyan1003/astrbot_plugin_kook_adapter)
 - [VoceChat](https://github.com/HikariFroya/astrbot_plugin_vocechat)
 - [Messages directs Bilibili](https://github.com/Hina-Chat/astrbot_plugin_bilibili_adapter)

README_ja.md

@@ -134,6 +134,7 @@ uv run main.py
 
 **コミュニティメンテナンス**
 
+- [Matrix](https://github.com/stevessr/astrbot_plugin_matrix_adapter)
 - [KOOK](https://github.com/wuyan1003/astrbot_plugin_kook_adapter)
 - [VoceChat](https://github.com/HikariFroya/astrbot_plugin_vocechat)
 - [Bilibili ダイレクトメッセージ](https://github.com/Hina-Chat/astrbot_plugin_bilibili_adapter)

README_ru.md

@@ -134,6 +134,7 @@ uv run main.py
 
 **Поддерживаемые сообществом**
 
+- [Matrix](https://github.com/stevessr/astrbot_plugin_matrix_adapter)
 - [KOOK](https://github.com/wuyan1003/astrbot_plugin_kook_adapter)
 - [VoceChat](https://github.com/HikariFroya/astrbot_plugin_vocechat)
 - [Личные сообщения Bilibili](https://github.com/Hina-Chat/astrbot_plugin_bilibili_adapter)

README_zh-TW.md

@@ -134,6 +134,7 @@ uv run main.py
 
 **社群維護**
 
+- [Matrix](https://github.com/stevessr/astrbot_plugin_matrix_adapter)
 - [KOOK](https://github.com/wuyan1003/astrbot_plugin_kook_adapter)
 - [VoceChat](https://github.com/HikariFroya/astrbot_plugin_vocechat)
 - [Bilibili 私訊](https://github.com/Hina-Chat/astrbot_plugin_bilibili_adapter)

astrbot/api/event/filter/__init__.py

@@ -21,6 +21,9 @@ from astrbot.core.star.register import (
 from astrbot.core.star.register import register_on_llm_request as on_llm_request
 from astrbot.core.star.register import register_on_llm_response as on_llm_response
 from astrbot.core.star.register import register_on_platform_loaded as on_platform_loaded
+from astrbot.core.star.register import (
+    register_on_waiting_llm_request as on_waiting_llm_request,
+)
 from astrbot.core.star.register import register_permission_type as permission_type
 from astrbot.core.star.register import (
     register_platform_adapter_type as platform_adapter_type,
@@ -46,6 +49,7 @@ __all__ = [
     "on_llm_request",
     "on_llm_response",
     "on_platform_loaded",
+    "on_waiting_llm_request",
     "permission_type",
     "platform_adapter_type",
     "regex",

astrbot/builtin_stars/astrbot/main.py

@@ -100,16 +100,8 @@ class Main(star.Star):
                 logger.error(f"ltm: {e}")
 
     @filter.on_llm_response()
-    async def inject_reasoning(self, event: AstrMessageEvent, resp: LLMResponse):
-        """在 LLM 响应后基于配置注入思考过程文本 / 在 LLM 响应后记录对话"""
-        umo = event.unified_msg_origin
-        cfg = self.context.get_config(umo).get("provider_settings", {})
-        show_reasoning = cfg.get("display_reasoning_text", False)
-        if show_reasoning and resp.reasoning_content:
-            resp.completion_text = (
-                f"🤔 思考: {resp.reasoning_content}\n\n{resp.completion_text}"
-            )
-
+    async def record_llm_resp_to_ltm(self, event: AstrMessageEvent, resp: LLMResponse):
+        """在 LLM 响应后记录对话"""
         if self.ltm and self.ltm_enabled(event):
             try:
                 await self.ltm.after_req_llm(event, resp)

astrbot/builtin_stars/astrbot/process_llm_request.py

@@ -7,6 +7,7 @@ from astrbot.api import logger, sp, star
 from astrbot.api.event import AstrMessageEvent
 from astrbot.api.message_components import Image, Reply
 from astrbot.api.provider import Provider, ProviderRequest
+from astrbot.core.agent.message import TextPart
 from astrbot.core.provider.func_tool_manager import ToolSet
 
 
@@ -85,7 +86,9 @@ class ProcessLLMRequest:
                 req.image_urls,
             )
             if caption:
-                req.prompt = f"(Image Caption: {caption})\n\n{req.prompt}"
+                req.extra_user_content_parts.append(
+                    TextPart(text=f"<image_caption>{caption}</image_caption>")
+                )
                 req.image_urls = []
         except Exception as e:
             logger.error(f"处理图片描述失败: {e}")
@@ -129,13 +132,14 @@ class ProcessLLMRequest:
             else:
                 req.prompt = prefix + req.prompt
 
+        # 收集系统提醒信息
+        system_parts = []
+
         # user identifier
         if cfg.get("identifier"):
             user_id = event.message_obj.sender.user_id
             user_nickname = event.message_obj.sender.nickname
-            req.prompt = (
-                f"\n[User ID: {user_id}, Nickname: {user_nickname}]\n{req.prompt}"
-            )
+            system_parts.append(f"User ID: {user_id}, Nickname: {user_nickname}")
 
         # group name identifier
         if cfg.get("group_name_display") and event.message_obj.group_id:
@@ -146,7 +150,7 @@ class ProcessLLMRequest:
                 return
             group_name = event.message_obj.group.group_name
             if group_name:
-                req.system_prompt += f"\nGroup name: {group_name}\n"
+                system_parts.append(f"Group name: {group_name}")
 
         # time info
         if cfg.get("datetime_system_prompt"):
@@ -162,7 +166,7 @@ class ProcessLLMRequest:
                 current_time = (
                     datetime.datetime.now().astimezone().strftime("%Y-%m-%d %H:%M (%Z)")
                 )
-            req.system_prompt += f"\nCurrent datetime: {current_time}\n"
+            system_parts.append(f"Current datetime: {current_time}")
 
         img_cap_prov_id: str = cfg.get("default_image_caption_provider_id") or ""
         if req.conversation:
@@ -225,10 +229,17 @@ class ProcessLLMRequest:
                 except BaseException as e:
                     logger.error(f"处理引用图片失败: {e}")
 
-            # 3. 将所有部分组合成文本并直接注入到当前消息中
+            # 3. 将所有部分组合成文本并添加到 extra_user_content_parts 中
             # 确保引用内容被正确的标签包裹
             quoted_content = "\n".join(content_parts)
             # 确保所有内容都在<Quoted Message>标签内
             quoted_text = f"<Quoted Message>\n{quoted_content}\n</Quoted Message>"
 
-            req.prompt = f"{quoted_text}\n\n{req.prompt}"
+            req.extra_user_content_parts.append(TextPart(text=quoted_text))
+
+        # 统一包裹所有系统提醒
+        if system_parts:
+            system_content = (
+                "<system_reminder>" + "\n".join(system_parts) + "</system_reminder>"
+            )
+            req.extra_user_content_parts.append(TextPart(text=system_content))

astrbot/builtin_stars/builtin_commands/commands/provider.py

@@ -184,7 +184,8 @@ class ProviderCommands:
                 event.set_result(MessageEventResult().message("请输入序号。"))
                 return
             if idx2 > len(self.context.get_all_tts_providers()) or idx2 < 1:
-                event.set_result(MessageEventResult().message("无效的序号。"))
+                event.set_result(MessageEventResult().message("无效的提供商序号。"))
+                return
             provider = self.context.get_all_tts_providers()[idx2 - 1]
             id_ = provider.meta().id
             await self.context.provider_manager.set_provider(
@@ -198,7 +199,8 @@ class ProviderCommands:
                 event.set_result(MessageEventResult().message("请输入序号。"))
                 return
             if idx2 > len(self.context.get_all_stt_providers()) or idx2 < 1:
-                event.set_result(MessageEventResult().message("无效的序号。"))
+                event.set_result(MessageEventResult().message("无效的提供商序号。"))
+                return
             provider = self.context.get_all_stt_providers()[idx2 - 1]
             id_ = provider.meta().id
             await self.context.provider_manager.set_provider(
@@ -209,8 +211,8 @@ class ProviderCommands:
             event.set_result(MessageEventResult().message(f"成功切换到 {id_}。"))
         elif isinstance(idx, int):
             if idx > len(self.context.get_all_providers()) or idx < 1:
-                event.set_result(MessageEventResult().message("无效的序号。"))
-
+                event.set_result(MessageEventResult().message("无效的提供商序号。"))
+                return
             provider = self.context.get_all_providers()[idx - 1]
             id_ = provider.meta().id
             await self.context.provider_manager.set_provider(

astrbot/builtin_stars/builtin_commands/commands/tts.py

@@ -14,13 +14,13 @@ class TTSCommand:
     async def tts(self, event: AstrMessageEvent):
         """开关文本转语音（会话级别）"""
         umo = event.unified_msg_origin
-        ses_tts = SessionServiceManager.is_tts_enabled_for_session(umo)
+        ses_tts = await SessionServiceManager.is_tts_enabled_for_session(umo)
         cfg = self.context.get_config(umo=umo)
         tts_enable = cfg["provider_tts_settings"]["enable"]
 
         # 切换状态
         new_status = not ses_tts
-        SessionServiceManager.set_tts_status_for_session(umo, new_status)
+        await SessionServiceManager.set_tts_status_for_session(umo, new_status)
 
         status_text = "已开启" if new_status else "已关闭"
 

astrbot/builtin_stars/python_interpreter/main.py

@@ -157,9 +157,8 @@ class Main(star.Star):
     async def is_docker_available(self) -> bool:
         """Check if docker is available"""
         try:
-            docker = aiodocker.Docker()
-            await docker.version()
-            await docker.close()
+            async with aiodocker.Docker() as docker:
+                await docker.version()
             return True
         except BaseException as e:
             logger.info(f"检查 Docker 可用性: {e}")
@@ -279,14 +278,14 @@ class Main(star.Star):
     @pi.command("repull")
     async def pi_repull(self, event: AstrMessageEvent):
         """重新拉取沙箱镜像"""
-        docker = aiodocker.Docker()
-        image_name = await self.get_image_name()
-        try:
-            await docker.images.get(image_name)
-            await docker.images.delete(image_name, force=True)
-        except aiodocker.exceptions.DockerError:
-            pass
-        await docker.images.pull(image_name)
+        async with aiodocker.Docker() as docker:
+            image_name = await self.get_image_name()
+            try:
+                await docker.images.get(image_name)
+                await docker.images.delete(image_name, force=True)
+            except aiodocker.exceptions.DockerError:
+                pass
+            await docker.images.pull(image_name)
         yield event.plain_result("重新拉取沙箱镜像成功。")
 
     @pi.command("file")
@@ -371,137 +370,137 @@ class Main(star.Star):
         obs = ""
         n = 5
 
-        for i in range(n):
-            if i > 0:
-                logger.info(f"Try {i + 1}/{n}")
+        async with aiodocker.Docker() as docker:
+            for i in range(n):
+                if i > 0:
+                    logger.info(f"Try {i + 1}/{n}")
 
-            PROMPT_ = PROMPT.format(
-                prompt=plain_text,
-                extra_input=extra_inputs,
-                extra_prompt=obs,
-            )
-            provider = self.context.get_using_provider()
-            llm_response = await provider.text_chat(
-                prompt=PROMPT_,
-                session_id=f"{event.session_id}_{magic_code}_{i!s}",
-            )
-
-            logger.debug(
-                "code interpreter llm gened code:" + llm_response.completion_text,
-            )
-
-            # 整理代码并保存
-            code_clean = await self.tidy_code(llm_response.completion_text)
-            with open(os.path.join(workplace_path, "exec.py"), "w") as f:
-                f.write(code_clean)
-
-            # 启动容器
-            docker = aiodocker.Docker()
-
-            # 检查有没有image
-            image_name = await self.get_image_name()
-            try:
-                await docker.images.get(image_name)
-            except aiodocker.exceptions.DockerError:
-                # 拉取镜像
-                logger.info(f"未找到沙箱镜像，正在尝试拉取 {image_name}...")
-                await docker.images.pull(image_name)
-
-            yield event.plain_result(
-                f"使用沙箱执行代码中，请稍等...(尝试次数: {i + 1}/{n})",
-            )
-
-            self.docker_host_astrbot_abs_path = self.config.get(
-                "docker_host_astrbot_abs_path",
-                "",
-            )
-            if self.docker_host_astrbot_abs_path:
-                host_shared = os.path.join(
-                    self.docker_host_astrbot_abs_path,
-                    self.shared_path,
+                PROMPT_ = PROMPT.format(
+                    prompt=plain_text,
+                    extra_input=extra_inputs,
+                    extra_prompt=obs,
                 )
-                host_output = os.path.join(
-                    self.docker_host_astrbot_abs_path,
-                    output_path,
-                )
-                host_workplace = os.path.join(
-                    self.docker_host_astrbot_abs_path,
-                    workplace_path,
+                provider = self.context.get_using_provider()
+                llm_response = await provider.text_chat(
+                    prompt=PROMPT_,
+                    session_id=f"{event.session_id}_{magic_code}_{i!s}",
                 )
 
-            else:
-                host_shared = os.path.abspath(self.shared_path)
-                host_output = os.path.abspath(output_path)
-                host_workplace = os.path.abspath(workplace_path)
+                logger.debug(
+                    "code interpreter llm gened code:" + llm_response.completion_text,
+                )
 
-            logger.debug(
-                f"host_shared: {host_shared}, host_output: {host_output}, host_workplace: {host_workplace}",
-            )
+                # 整理代码并保存
+                code_clean = await self.tidy_code(llm_response.completion_text)
+                with open(os.path.join(workplace_path, "exec.py"), "w") as f:
+                    f.write(code_clean)
 
-            container = await docker.containers.run(
-                {
-                    "Image": image_name,
-                    "Cmd": ["python", "exec.py"],
-                    "Memory": 512 * 1024 * 1024,
-                    "NanoCPUs": 1000000000,
-                    "HostConfig": {
-                        "Binds": [
-                            f"{host_shared}:/astrbot_sandbox/shared:ro",
-                            f"{host_output}:/astrbot_sandbox/output:rw",
-                            f"{host_workplace}:/astrbot_sandbox:rw",
-                        ],
-                    },
-                    "Env": [f"MAGIC_CODE={magic_code}"],
-                    "AutoRemove": True,
-                },
-            )
+                # 检查有没有image
+                image_name = await self.get_image_name()
+                try:
+                    await docker.images.get(image_name)
+                except aiodocker.exceptions.DockerError:
+                    # 拉取镜像
+                    logger.info(f"未找到沙箱镜像，正在尝试拉取 {image_name}...")
+                    await docker.images.pull(image_name)
 
-            logger.debug(f"Container {container.id} created.")
-            logs = await self.run_container(container)
+                yield event.plain_result(
+                    f"使用沙箱执行代码中，请稍等...(尝试次数: {i + 1}/{n})",
+                )
 
-            logger.debug(f"Container {container.id} finished.")
-            logger.debug(f"Container {container.id} logs: {logs}")
-
-            # 发送结果
-            pattern = r"\[ASTRBOT_(TEXT|IMAGE|FILE)_OUTPUT#\w+\]: (.*)"
-            ok = False
-            traceback = ""
-            for idx, log in enumerate(logs):
-                match = re.match(pattern, log)
-                if match:
-                    ok = True
-                    if match.group(1) == "TEXT":
-                        yield event.plain_result(match.group(2))
-                    elif match.group(1) == "IMAGE":
-                        image_path = os.path.join(workplace_path, match.group(2))
-                        logger.debug(f"Sending image: {image_path}")
-                        yield event.image_result(image_path)
-                    elif match.group(1) == "FILE":
-                        file_path = os.path.join(workplace_path, match.group(2))
-                        # logger.debug(f"Sending file: {file_path}")
-                        # file_s3_url = await self.file_upload(file_path)
-                        # logger.info(f"文件上传到 AstrBot 云节点: {file_s3_url}")
-                        file_name = os.path.basename(file_path)
-                        chain: list[BaseMessageComponent] = [
-                            File(name=file_name, file=file_path)
-                        ]
-                        yield event.set_result(MessageEventResult(chain=chain))
-
-                elif "Traceback (most recent call last)" in log or "[Error]: " in log:
-                    traceback = "\n".join(logs[idx:])
-
-            if not ok:
-                if traceback:
-                    obs = f"## Observation \n When execute the code: ```python\n{code_clean}\n```\n\n Error occurred:\n\n{traceback}\n Need to improve/fix the code."
-                else:
-                    logger.warning(
-                        f"未从沙箱输出中捕获到合法的输出。沙箱输出日志: {logs}",
+                self.docker_host_astrbot_abs_path = self.config.get(
+                    "docker_host_astrbot_abs_path",
+                    "",
+                )
+                if self.docker_host_astrbot_abs_path:
+                    host_shared = os.path.join(
+                        self.docker_host_astrbot_abs_path,
+                        self.shared_path,
                     )
-                    break
-            else:
-                # 成功了
-                self.user_file_msg_buffer.pop(event.get_session_id())
-                return
+                    host_output = os.path.join(
+                        self.docker_host_astrbot_abs_path,
+                        output_path,
+                    )
+                    host_workplace = os.path.join(
+                        self.docker_host_astrbot_abs_path,
+                        workplace_path,
+                    )
+
+                else:
+                    host_shared = os.path.abspath(self.shared_path)
+                    host_output = os.path.abspath(output_path)
+                    host_workplace = os.path.abspath(workplace_path)
+
+                logger.debug(
+                    f"host_shared: {host_shared}, host_output: {host_output}, host_workplace: {host_workplace}",
+                )
+
+                container = await docker.containers.run(
+                    {
+                        "Image": image_name,
+                        "Cmd": ["python", "exec.py"],
+                        "Memory": 512 * 1024 * 1024,
+                        "NanoCPUs": 1000000000,
+                        "HostConfig": {
+                            "Binds": [
+                                f"{host_shared}:/astrbot_sandbox/shared:ro",
+                                f"{host_output}:/astrbot_sandbox/output:rw",
+                                f"{host_workplace}:/astrbot_sandbox:rw",
+                            ],
+                        },
+                        "Env": [f"MAGIC_CODE={magic_code}"],
+                        "AutoRemove": True,
+                    },
+                )
+
+                logger.debug(f"Container {container.id} created.")
+                logs = await self.run_container(container)
+
+                logger.debug(f"Container {container.id} finished.")
+                logger.debug(f"Container {container.id} logs: {logs}")
+
+                # 发送结果
+                pattern = r"\[ASTRBOT_(TEXT|IMAGE|FILE)_OUTPUT#\w+\]: (.*)"
+                ok = False
+                traceback = ""
+                for idx, log in enumerate(logs):
+                    match = re.match(pattern, log)
+                    if match:
+                        ok = True
+                        if match.group(1) == "TEXT":
+                            yield event.plain_result(match.group(2))
+                        elif match.group(1) == "IMAGE":
+                            image_path = os.path.join(workplace_path, match.group(2))
+                            logger.debug(f"Sending image: {image_path}")
+                            yield event.image_result(image_path)
+                        elif match.group(1) == "FILE":
+                            file_path = os.path.join(workplace_path, match.group(2))
+                            # logger.debug(f"Sending file: {file_path}")
+                            # file_s3_url = await self.file_upload(file_path)
+                            # logger.info(f"文件上传到 AstrBot 云节点: {file_s3_url}")
+                            file_name = os.path.basename(file_path)
+                            chain: list[BaseMessageComponent] = [
+                                File(name=file_name, file=file_path)
+                            ]
+                            yield event.set_result(MessageEventResult(chain=chain))
+
+                    elif (
+                        "Traceback (most recent call last)" in log or "[Error]: " in log
+                    ):
+                        traceback = "\n".join(logs[idx:])
+
+                if not ok:
+                    if traceback:
+                        obs = f"## Observation \n When execute the code: ```python\n{code_clean}\n```\n\n Error occurred:\n\n{traceback}\n Need to improve/fix the code."
+                    else:
+                        logger.warning(
+                            f"未从沙箱输出中捕获到合法的输出。沙箱输出日志: {logs}",
+                        )
+                        break
+                else:
+                    # 成功了
+                    self.user_file_msg_buffer.pop(event.get_session_id())
+                    return
 
         yield event.plain_result(
             "经过多次尝试后，未从沙箱输出中捕获到合法的输出，请更换问法或者查看日志。",

astrbot/cli/__init__.py

@@ -1 +1 @@
-__version__ = "4.10.0"
+__version__ = "4.11.0"

astrbot/core/agent/context/compressor.py

@@ -0,0 +1,243 @@
+from typing import TYPE_CHECKING, Protocol, runtime_checkable
+
+from ..message import Message
+
+if TYPE_CHECKING:
+    from astrbot import logger
+else:
+    try:
+        from astrbot import logger
+    except ImportError:
+        import logging
+
+        logger = logging.getLogger("astrbot")
+
+if TYPE_CHECKING:
+    from astrbot.core.provider.provider import Provider
+
+from ..context.truncator import ContextTruncator
+
+
+@runtime_checkable
+class ContextCompressor(Protocol):
+    """
+    Protocol for context compressors.
+    Provides an interface for compressing message lists.
+    """
+
+    def should_compress(
+        self, messages: list[Message], current_tokens: int, max_tokens: int
+    ) -> bool:
+        """Check if compression is needed.
+
+        Args:
+            messages: The message list to evaluate.
+            current_tokens: The current token count.
+            max_tokens: The maximum allowed tokens for the model.
+
+        Returns:
+            True if compression is needed, False otherwise.
+        """
+        ...
+
+    async def __call__(self, messages: list[Message]) -> list[Message]:
+        """Compress the message list.
+
+        Args:
+            messages: The original message list.
+
+        Returns:
+            The compressed message list.
+        """
+        ...
+
+
+class TruncateByTurnsCompressor:
+    """Truncate by turns compressor implementation.
+    Truncates the message list by removing older turns.
+    """
+
+    def __init__(self, truncate_turns: int = 1, compression_threshold: float = 0.82):
+        """Initialize the truncate by turns compressor.
+
+        Args:
+            truncate_turns: The number of turns to remove when truncating (default: 1).
+            compression_threshold: The compression trigger threshold (default: 0.82).
+        """
+        self.truncate_turns = truncate_turns
+        self.compression_threshold = compression_threshold
+
+    def should_compress(
+        self, messages: list[Message], current_tokens: int, max_tokens: int
+    ) -> bool:
+        """Check if compression is needed.
+
+        Args:
+            messages: The message list to evaluate.
+            current_tokens: The current token count.
+            max_tokens: The maximum allowed tokens.
+
+        Returns:
+            True if compression is needed, False otherwise.
+        """
+        if max_tokens <= 0 or current_tokens <= 0:
+            return False
+        usage_rate = current_tokens / max_tokens
+        return usage_rate > self.compression_threshold
+
+    async def __call__(self, messages: list[Message]) -> list[Message]:
+        truncator = ContextTruncator()
+        truncated_messages = truncator.truncate_by_dropping_oldest_turns(
+            messages,
+            drop_turns=self.truncate_turns,
+        )
+        return truncated_messages
+
+
+def split_history(
+    messages: list[Message], keep_recent: int
+) -> tuple[list[Message], list[Message], list[Message]]:
+    """Split the message list into system messages, messages to summarize, and recent messages.
+
+    Ensures that the split point is between complete user-assistant pairs to maintain conversation flow.
+
+    Args:
+        messages: The original message list.
+        keep_recent: The number of latest messages to keep.
+
+    Returns:
+        tuple: (system_messages, messages_to_summarize, recent_messages)
+    """
+    # keep the system messages
+    first_non_system = 0
+    for i, msg in enumerate(messages):
+        if msg.role != "system":
+            first_non_system = i
+            break
+
+    system_messages = messages[:first_non_system]
+    non_system_messages = messages[first_non_system:]
+
+    if len(non_system_messages) <= keep_recent:
+        return system_messages, [], non_system_messages
+
+    # Find the split point, ensuring recent_messages starts with a user message
+    # This maintains complete conversation turns
+    split_index = len(non_system_messages) - keep_recent
+
+    # Search backward from split_index to find the first user message
+    # This ensures recent_messages starts with a user message (complete turn)
+    while split_index > 0 and non_system_messages[split_index].role != "user":
+        # TODO: +=1 or -=1 ? calculate by tokens
+        split_index -= 1
+
+    # If we couldn't find a user message, keep all messages as recent
+    if split_index == 0:
+        return system_messages, [], non_system_messages
+
+    messages_to_summarize = non_system_messages[:split_index]
+    recent_messages = non_system_messages[split_index:]
+
+    return system_messages, messages_to_summarize, recent_messages
+
+
+class LLMSummaryCompressor:
+    """LLM-based summary compressor.
+    Uses LLM to summarize the old conversation history, keeping the latest messages.
+    """
+
+    def __init__(
+        self,
+        provider: "Provider",
+        keep_recent: int = 4,
+        instruction_text: str | None = None,
+        compression_threshold: float = 0.82,
+    ):
+        """Initialize the LLM summary compressor.
+
+        Args:
+            provider: The LLM provider instance.
+            keep_recent: The number of latest messages to keep (default: 4).
+            instruction_text: Custom instruction for summary generation.
+            compression_threshold: The compression trigger threshold (default: 0.82).
+        """
+        self.provider = provider
+        self.keep_recent = keep_recent
+        self.compression_threshold = compression_threshold
+
+        self.instruction_text = instruction_text or (
+            "Based on our full conversation history, produce a concise summary of key takeaways and/or project progress.\n"
+            "1. Systematically cover all core topics discussed and the final conclusion/outcome for each; clearly highlight the latest primary focus.\n"
+            "2. If any tools were used, summarize tool usage (total call count) and extract the most valuable insights from tool outputs.\n"
+            "3. If there was an initial user goal, state it first and describe the current progress/status.\n"
+            "4. Write the summary in the user's language.\n"
+        )
+
+    def should_compress(
+        self, messages: list[Message], current_tokens: int, max_tokens: int
+    ) -> bool:
+        """Check if compression is needed.
+
+        Args:
+            messages: The message list to evaluate.
+            current_tokens: The current token count.
+            max_tokens: The maximum allowed tokens.
+
+        Returns:
+            True if compression is needed, False otherwise.
+        """
+        if max_tokens <= 0 or current_tokens <= 0:
+            return False
+        usage_rate = current_tokens / max_tokens
+        return usage_rate > self.compression_threshold
+
+    async def __call__(self, messages: list[Message]) -> list[Message]:
+        """Use LLM to generate a summary of the conversation history.
+
+        Process:
+        1. Divide messages: keep the system message and the latest N messages.
+        2. Send the old messages + the instruction message to the LLM.
+        3. Reconstruct the message list: [system message, summary message, latest messages].
+        """
+        if len(messages) <= self.keep_recent + 1:
+            return messages
+
+        system_messages, messages_to_summarize, recent_messages = split_history(
+            messages, self.keep_recent
+        )
+
+        if not messages_to_summarize:
+            return messages
+
+        # build payload
+        instruction_message = Message(role="user", content=self.instruction_text)
+        llm_payload = messages_to_summarize + [instruction_message]
+
+        # generate summary
+        try:
+            response = await self.provider.text_chat(contexts=llm_payload)
+            summary_content = response.completion_text
+        except Exception as e:
+            logger.error(f"Failed to generate summary: {e}")
+            return messages
+
+        # build result
+        result = []
+        result.extend(system_messages)
+
+        result.append(
+            Message(
+                role="user",
+                content=f"Our previous history conversation summary: {summary_content}",
+            )
+        )
+        result.append(
+            Message(
+                role="assistant",
+                content="Acknowledged the summary of our previous conversation history.",
+            )
+        )
+
+        result.extend(recent_messages)
+
+        return result

astrbot/core/agent/context/config.py

@@ -0,0 +1,35 @@
+from dataclasses import dataclass
+from typing import TYPE_CHECKING
+
+from .compressor import ContextCompressor
+from .token_counter import TokenCounter
+
+if TYPE_CHECKING:
+    from astrbot.core.provider.provider import Provider
+
+
+@dataclass
+class ContextConfig:
+    """Context configuration class."""
+
+    max_context_tokens: int = 0
+    """Maximum number of context tokens. <= 0 means no limit."""
+    enforce_max_turns: int = -1  # -1 means no limit
+    """Maximum number of conversation turns to keep. -1 means no limit. Executed before compression."""
+    truncate_turns: int = 1
+    """Number of conversation turns to discard at once when truncation is triggered.
+    Two processes will use this value:
+
+    1. Enforce max turns truncation.
+    2. Truncation by turns compression strategy.
+    """
+    llm_compress_instruction: str | None = None
+    """Instruction prompt for LLM-based compression."""
+    llm_compress_keep_recent: int = 0
+    """Number of recent messages to keep during LLM-based compression."""
+    llm_compress_provider: "Provider | None" = None
+    """LLM provider used for compression tasks. If None, truncation strategy is used."""
+    custom_token_counter: TokenCounter | None = None
+    """Custom token counting method. If None, the default method is used."""
+    custom_compressor: ContextCompressor | None = None
+    """Custom context compression method. If None, the default method is used."""

astrbot/core/agent/context/manager.py

@@ -0,0 +1,120 @@
+from astrbot import logger
+
+from ..message import Message
+from .compressor import LLMSummaryCompressor, TruncateByTurnsCompressor
+from .config import ContextConfig
+from .token_counter import EstimateTokenCounter
+from .truncator import ContextTruncator
+
+
+class ContextManager:
+    """Context compression manager."""
+
+    def __init__(
+        self,
+        config: ContextConfig,
+    ):
+        """Initialize the context manager.
+
+        There are two strategies to handle context limit reached:
+        1. Truncate by turns: remove older messages by turns.
+        2. LLM-based compression: use LLM to summarize old messages.
+
+        Args:
+            config: The context configuration.
+        """
+        self.config = config
+
+        self.token_counter = config.custom_token_counter or EstimateTokenCounter()
+        self.truncator = ContextTruncator()
+
+        if config.custom_compressor:
+            self.compressor = config.custom_compressor
+        elif config.llm_compress_provider:
+            self.compressor = LLMSummaryCompressor(
+                provider=config.llm_compress_provider,
+                keep_recent=config.llm_compress_keep_recent,
+                instruction_text=config.llm_compress_instruction,
+            )
+        else:
+            self.compressor = TruncateByTurnsCompressor(
+                truncate_turns=config.truncate_turns
+            )
+
+    async def process(
+        self, messages: list[Message], trusted_token_usage: int = 0
+    ) -> list[Message]:
+        """Process the messages.
+
+        Args:
+            messages: The original message list.
+
+        Returns:
+            The processed message list.
+        """
+        try:
+            result = messages
+
+            # 1. 基于轮次的截断 (Enforce max turns)
+            if self.config.enforce_max_turns != -1:
+                result = self.truncator.truncate_by_turns(
+                    result,
+                    keep_most_recent_turns=self.config.enforce_max_turns,
+                    drop_turns=self.config.truncate_turns,
+                )
+
+            # 2. 基于 token 的压缩
+            if self.config.max_context_tokens > 0:
+                total_tokens = self.token_counter.count_tokens(
+                    result, trusted_token_usage
+                )
+
+                if self.compressor.should_compress(
+                    result, total_tokens, self.config.max_context_tokens
+                ):
+                    result = await self._run_compression(result, total_tokens)
+
+            return result
+        except Exception as e:
+            logger.error(f"Error during context processing: {e}", exc_info=True)
+            return messages
+
+    async def _run_compression(
+        self, messages: list[Message], prev_tokens: int
+    ) -> list[Message]:
+        """
+        Compress/truncate the messages.
+
+        Args:
+            messages: The original message list.
+            prev_tokens: The token count before compression.
+
+        Returns:
+            The compressed/truncated message list.
+        """
+        logger.debug("Compress triggered, starting compression...")
+
+        messages = await self.compressor(messages)
+
+        # double check
+        tokens_after_summary = self.token_counter.count_tokens(messages)
+
+        # calculate compress rate
+        compress_rate = (tokens_after_summary / self.config.max_context_tokens) * 100
+        logger.info(
+            f"Compress completed."
+            f" {prev_tokens} -> {tokens_after_summary} tokens,"
+            f" compression rate: {compress_rate:.2f}%.",
+        )
+
+        # last check
+        if self.compressor.should_compress(
+            messages, tokens_after_summary, self.config.max_context_tokens
+        ):
+            logger.info(
+                "Context still exceeds max tokens after compression, applying halving truncation..."
+            )
+            # still need compress, truncate by half
+            messages = self.truncator.truncate_by_halving(messages)
+
+        return messages

astrbot/core/agent/context/token_counter.py

@@ -0,0 +1,64 @@
+import json
+from typing import Protocol, runtime_checkable
+
+from ..message import Message, TextPart
+
+
+@runtime_checkable
+class TokenCounter(Protocol):
+    """
+    Protocol for token counters.
+    Provides an interface for counting tokens in message lists.
+    """
+
+    def count_tokens(
+        self, messages: list[Message], trusted_token_usage: int = 0
+    ) -> int:
+        """Count the total tokens in the message list.
+
+        Args:
+            messages: The message list.
+            trusted_token_usage: The total token usage that LLM API returned.
+                For some cases, this value is more accurate.
+                But some API does not return it, so the value defaults to 0.
+
+        Returns:
+            The total token count.
+        """
+        ...
+
+
+class EstimateTokenCounter:
+    """Estimate token counter implementation.
+    Provides a simple estimation of token count based on character types.
+    """
+
+    def count_tokens(
+        self, messages: list[Message], trusted_token_usage: int = 0
+    ) -> int:
+        if trusted_token_usage > 0:
+            return trusted_token_usage
+
+        total = 0
+        for msg in messages:
+            content = msg.content
+            if isinstance(content, str):
+                total += self._estimate_tokens(content)
+            elif isinstance(content, list):
+                # 处理多模态内容
+                for part in content:
+                    if isinstance(part, TextPart):
+                        total += self._estimate_tokens(part.text)
+
+            # 处理 Tool Calls
+            if msg.tool_calls:
+                for tc in msg.tool_calls:
+                    tc_str = json.dumps(tc if isinstance(tc, dict) else tc.model_dump())
+                    total += self._estimate_tokens(tc_str)
+
+        return total
+
+    def _estimate_tokens(self, text: str) -> int:
+        chinese_count = len([c for c in text if "\u4e00" <= c <= "\u9fff"])
+        other_count = len(text) - chinese_count
+        return int(chinese_count * 0.6 + other_count * 0.3)

astrbot/core/agent/context/truncator.py

@@ -0,0 +1,141 @@
+from ..message import Message
+
+
+class ContextTruncator:
+    """Context truncator."""
+
+    def fix_messages(self, messages: list[Message]) -> list[Message]:
+        fixed_messages = []
+        for message in messages:
+            if message.role == "tool":
+                # tool block 前面必须要有 user 和 assistant block
+                if len(fixed_messages) < 2:
+                    # 这种情况可能是上下文被截断导致的
+                    # 我们直接将之前的上下文都清空
+                    fixed_messages = []
+                else:
+                    fixed_messages.append(message)
+            else:
+                fixed_messages.append(message)
+        return fixed_messages
+
+    def truncate_by_turns(
+        self,
+        messages: list[Message],
+        keep_most_recent_turns: int,
+        drop_turns: int = 1,
+    ) -> list[Message]:
+        """截断上下文列表，确保不超过最大长度。
+        一个 turn 包含一个 user 消息和一个 assistant 消息。
+        这个方法会保证截断后的上下文列表符合 OpenAI 的上下文格式。
+
+        Args:
+            messages: 上下文列表
+            keep_most_recent_turns: 保留最近的对话轮数
+            drop_turns: 一次性丢弃的对话轮数
+
+        Returns:
+            截断后的上下文列表
+        """
+        if keep_most_recent_turns == -1:
+            return messages
+
+        first_non_system = 0
+        for i, msg in enumerate(messages):
+            if msg.role != "system":
+                first_non_system = i
+                break
+
+        system_messages = messages[:first_non_system]
+        non_system_messages = messages[first_non_system:]
+
+        if len(non_system_messages) // 2 <= keep_most_recent_turns:
+            return messages
+
+        num_to_keep = keep_most_recent_turns - drop_turns + 1
+        if num_to_keep <= 0:
+            truncated_contexts = []
+        else:
+            truncated_contexts = non_system_messages[-num_to_keep * 2 :]
+
+        # 找到第一个 role 为 user 的索引，确保上下文格式正确
+        index = next(
+            (i for i, item in enumerate(truncated_contexts) if item.role == "user"),
+            None,
+        )
+        if index is not None and index > 0:
+            truncated_contexts = truncated_contexts[index:]
+
+        result = system_messages + truncated_contexts
+
+        return self.fix_messages(result)
+
+    def truncate_by_dropping_oldest_turns(
+        self,
+        messages: list[Message],
+        drop_turns: int = 1,
+    ) -> list[Message]:
+        """丢弃最旧的 N 个对话轮次。"""
+        if drop_turns <= 0:
+            return messages
+
+        first_non_system = 0
+        for i, msg in enumerate(messages):
+            if msg.role != "system":
+                first_non_system = i
+                break
+
+        system_messages = messages[:first_non_system]
+        non_system_messages = messages[first_non_system:]
+
+        if len(non_system_messages) // 2 <= drop_turns:
+            truncated_non_system = []
+        else:
+            truncated_non_system = non_system_messages[drop_turns * 2 :]
+
+        index = next(
+            (i for i, item in enumerate(truncated_non_system) if item.role == "user"),
+            None,
+        )
+        if index is not None:
+            truncated_non_system = truncated_non_system[index:]
+        elif truncated_non_system:
+            truncated_non_system = []
+
+        result = system_messages + truncated_non_system
+
+        return self.fix_messages(result)
+
+    def truncate_by_halving(
+        self,
+        messages: list[Message],
+    ) -> list[Message]:
+        """对半砍策略，删除 50% 的消息"""
+        if len(messages) <= 2:
+            return messages
+
+        first_non_system = 0
+        for i, msg in enumerate(messages):
+            if msg.role != "system":
+                first_non_system = i
+                break
+
+        system_messages = messages[:first_non_system]
+        non_system_messages = messages[first_non_system:]
+
+        messages_to_delete = len(non_system_messages) // 2
+        if messages_to_delete == 0:
+            return messages
+
+        truncated_non_system = non_system_messages[messages_to_delete:]
+
+        index = next(
+            (i for i, item in enumerate(truncated_non_system) if item.role == "user"),
+            None,
+        )
+        if index is not None:
+            truncated_non_system = truncated_non_system[index:]
+
+        result = system_messages + truncated_non_system
+
+        return self.fix_messages(result)

astrbot/core/agent/message.py

@@ -12,7 +12,7 @@ class ContentPart(BaseModel):
 
     __content_part_registry: ClassVar[dict[str, type["ContentPart"]]] = {}
 
-    type: str
+    type: Literal["text", "think", "image_url", "audio_url"]
 
     def __init_subclass__(cls, **kwargs: Any) -> None:
         super().__init_subclass__(**kwargs)
@@ -63,6 +63,28 @@ class TextPart(ContentPart):
     text: str
 
 
+class ThinkPart(ContentPart):
+    """
+    >>> ThinkPart(think="I think I need to think about this.").model_dump()
+    {'type': 'think', 'think': 'I think I need to think about this.', 'encrypted': None}
+    """
+
+    type: str = "think"
+    think: str
+    encrypted: str | None = None
+    """Encrypted thinking content, or signature."""
+
+    def merge_in_place(self, other: Any) -> bool:
+        if not isinstance(other, ThinkPart):
+            return False
+        if self.encrypted:
+            return False
+        self.think += other.think
+        if other.encrypted:
+            self.encrypted = other.encrypted
+        return True
+
+
 class ImageURLPart(ContentPart):
     """
     >>> ImageURLPart(image_url="http://example.com/image.jpg").model_dump()
@@ -169,6 +191,15 @@ class Message(BaseModel):
             )
         return self
 
+    @model_serializer(mode="wrap")
+    def serialize(self, handler):
+        data = handler(self)
+        if self.tool_calls is None:
+            data.pop("tool_calls", None)
+        if self.tool_call_id is None:
+            data.pop("tool_call_id", None)
+        return data
+
 
 class AssistantMessageSegment(Message):
     """A message segment from the assistant."""

astrbot/core/agent/runners/tool_loop_agent_runner.py

@@ -13,6 +13,7 @@ from mcp.types import (
 )
 
 from astrbot import logger
+from astrbot.core.agent.message import TextPart, ThinkPart
 from astrbot.core.message.components import Json
 from astrbot.core.message.message_event_result import (
     MessageChain,
@@ -24,6 +25,10 @@ from astrbot.core.provider.entities import (
 )
 from astrbot.core.provider.provider import Provider
 
+from ..context.compressor import ContextCompressor
+from ..context.config import ContextConfig
+from ..context.manager import ContextManager
+from ..context.token_counter import TokenCounter
 from ..hooks import BaseAgentRunHooks
 from ..message import AssistantMessageSegment, Message, ToolCallMessageSegment
 from ..response import AgentResponseData, AgentStats
@@ -46,10 +51,47 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]):
         run_context: ContextWrapper[TContext],
         tool_executor: BaseFunctionToolExecutor[TContext],
         agent_hooks: BaseAgentRunHooks[TContext],
+        streaming: bool = False,
+        # enforce max turns, will discard older turns when exceeded BEFORE compression
+        # -1 means no limit
+        enforce_max_turns: int = -1,
+        # llm compressor
+        llm_compress_instruction: str | None = None,
+        llm_compress_keep_recent: int = 0,
+        llm_compress_provider: Provider | None = None,
+        # truncate by turns compressor
+        truncate_turns: int = 1,
+        # customize
+        custom_token_counter: TokenCounter | None = None,
+        custom_compressor: ContextCompressor | None = None,
         **kwargs: T.Any,
     ) -> None:
         self.req = request
-        self.streaming = kwargs.get("streaming", False)
+        self.streaming = streaming
+        self.enforce_max_turns = enforce_max_turns
+        self.llm_compress_instruction = llm_compress_instruction
+        self.llm_compress_keep_recent = llm_compress_keep_recent
+        self.llm_compress_provider = llm_compress_provider
+        self.truncate_turns = truncate_turns
+        self.custom_token_counter = custom_token_counter
+        self.custom_compressor = custom_compressor
+        # we will do compress when:
+        # 1. before requesting LLM
+        # TODO: 2. after LLM output a tool call
+        self.context_config = ContextConfig(
+            # <=0 will never do compress
+            max_context_tokens=provider.provider_config.get("max_context_tokens", 0),
+            # enforce max turns before compression
+            enforce_max_turns=self.enforce_max_turns,
+            truncate_turns=self.truncate_turns,
+            llm_compress_instruction=self.llm_compress_instruction,
+            llm_compress_keep_recent=self.llm_compress_keep_recent,
+            llm_compress_provider=self.llm_compress_provider,
+            custom_token_counter=self.custom_token_counter,
+            custom_compressor=self.custom_compressor,
+        )
+        self.context_manager = ContextManager(self.context_config)
+
         self.provider = provider
         self.final_llm_resp = None
         self._state = AgentState.IDLE
@@ -77,10 +119,11 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]):
     async def _iter_llm_responses(self) -> T.AsyncGenerator[LLMResponse, None]:
         """Yields chunks *and* a final LLMResponse."""
         payload = {
-            "contexts": self.run_context.messages,
+            "contexts": self.run_context.messages,  # list[Message]
             "func_tool": self.req.func_tool,
             "model": self.req.model,  # NOTE: in fact, this arg is None in most cases
             "session_id": self.req.session_id,
+            "extra_user_content_parts": self.req.extra_user_content_parts,  # list[ContentPart]
         }
 
         if self.streaming:
@@ -108,6 +151,12 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]):
         self._transition_state(AgentState.RUNNING)
         llm_resp_result = None
 
+        # do truncate and compress
+        token_usage = self.req.conversation.token_usage if self.req.conversation else 0
+        self.run_context.messages = await self.context_manager.process(
+            self.run_context.messages, trusted_token_usage=token_usage
+        )
+
         async for llm_response in self._iter_llm_responses():
             if llm_response.is_chunk:
                 # update ttft
@@ -168,13 +217,20 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]):
             self.final_llm_resp = llm_resp
             self._transition_state(AgentState.DONE)
             self.stats.end_time = time.time()
+
             # record the final assistant message
-            self.run_context.messages.append(
-                Message(
-                    role="assistant",
-                    content=llm_resp.completion_text or "*No response*",
-                ),
-            )
+            parts = []
+            if llm_resp.reasoning_content or llm_resp.reasoning_signature:
+                parts.append(
+                    ThinkPart(
+                        think=llm_resp.reasoning_content,
+                        encrypted=llm_resp.reasoning_signature,
+                    )
+                )
+            parts.append(TextPart(text=llm_resp.completion_text or "*No response*"))
+            self.run_context.messages.append(Message(role="assistant", content=parts))
+
+            # call the on_agent_done hook
             try:
                 await self.agent_hooks.on_agent_done(self.run_context, llm_resp)
             except Exception as e:
@@ -213,10 +269,19 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]):
                         data=AgentResponseData(chain=result),
                     )
             # 将结果添加到上下文中
+            parts = []
+            if llm_resp.reasoning_content or llm_resp.reasoning_signature:
+                parts.append(
+                    ThinkPart(
+                        think=llm_resp.reasoning_content,
+                        encrypted=llm_resp.reasoning_signature,
+                    )
+                )
+            parts.append(TextPart(text=llm_resp.completion_text or "*No response*"))
             tool_calls_result = ToolCallsResult(
                 tool_calls_info=AssistantMessageSegment(
                     tool_calls=llm_resp.to_openai_to_calls_model(),
-                    content=llm_resp.completion_text,
+                    content=parts,
                 ),
                 tool_calls_result=tool_call_result_blocks,
             )

astrbot/core/astr_agent_hooks.py

@@ -13,6 +13,12 @@ from astrbot.core.star.star_handler import EventType
 class MainAgentHooks(BaseAgentRunHooks[AstrAgentContext]):
     async def on_agent_done(self, run_context, llm_response):
         # 执行事件钩子
+        if llm_response and llm_response.reasoning_content:
+            # we will use this in result_decorate stage to inject reasoning content to chain
+            run_context.context.event.set_extra(
+                "_llm_reasoning_content", llm_response.reasoning_content
+            )
+
         await call_event_hook(
             run_context.context.event,
             EventType.OnLLMResponseEvent,

astrbot/core/backup/__init__.py

@@ -0,0 +1,26 @@
+"""AstrBot 备份与恢复模块
+
+提供数据导出和导入功能，支持用户在服务器迁移时一键备份和恢复所有数据。
+"""
+
+# 从 constants 模块导入共享常量
+from .constants import (
+    BACKUP_MANIFEST_VERSION,
+    KB_METADATA_MODELS,
+    MAIN_DB_MODELS,
+    get_backup_directories,
+)
+
+# 导入导出器和导入器
+from .exporter import AstrBotExporter
+from .importer import AstrBotImporter, ImportPreCheckResult
+
+__all__ = [
+    "AstrBotExporter",
+    "AstrBotImporter",
+    "ImportPreCheckResult",
+    "MAIN_DB_MODELS",
+    "KB_METADATA_MODELS",
+    "get_backup_directories",
+    "BACKUP_MANIFEST_VERSION",
+]

astrbot/core/backup/constants.py

@@ -0,0 +1,77 @@
+"""AstrBot 备份模块共享常量
+
+此文件定义了导出器和导入器共享的常量，确保两端配置一致。
+"""
+
+from sqlmodel import SQLModel
+
+from astrbot.core.db.po import (
+    Attachment,
+    CommandConfig,
+    CommandConflict,
+    ConversationV2,
+    Persona,
+    PlatformMessageHistory,
+    PlatformSession,
+    PlatformStat,
+    Preference,
+)
+from astrbot.core.knowledge_base.models import (
+    KBDocument,
+    KBMedia,
+    KnowledgeBase,
+)
+from astrbot.core.utils.astrbot_path import (
+    get_astrbot_config_path,
+    get_astrbot_plugin_data_path,
+    get_astrbot_plugin_path,
+    get_astrbot_t2i_templates_path,
+    get_astrbot_temp_path,
+    get_astrbot_webchat_path,
+)
+
+# ============================================================
+# 共享常量 - 确保导出和导入端配置一致
+# ============================================================
+
+# 主数据库模型类映射
+MAIN_DB_MODELS: dict[str, type[SQLModel]] = {
+    "platform_stats": PlatformStat,
+    "conversations": ConversationV2,
+    "personas": Persona,
+    "preferences": Preference,
+    "platform_message_history": PlatformMessageHistory,
+    "platform_sessions": PlatformSession,
+    "attachments": Attachment,
+    "command_configs": CommandConfig,
+    "command_conflicts": CommandConflict,
+}
+
+# 知识库元数据模型类映射
+KB_METADATA_MODELS: dict[str, type[SQLModel]] = {
+    "knowledge_bases": KnowledgeBase,
+    "kb_documents": KBDocument,
+    "kb_media": KBMedia,
+}
+
+
+def get_backup_directories() -> dict[str, str]:
+    """获取需要备份的目录列表
+
+    使用 astrbot_path 模块动态获取路径，支持通过环境变量 ASTRBOT_ROOT 自定义根目录。
+
+    Returns:
+        dict: 键为备份文件中的目录名称，值为目录的绝对路径
+    """
+    return {
+        "plugins": get_astrbot_plugin_path(),  # 插件本体
+        "plugin_data": get_astrbot_plugin_data_path(),  # 插件数据
+        "config": get_astrbot_config_path(),  # 配置目录
+        "t2i_templates": get_astrbot_t2i_templates_path(),  # T2I 模板
+        "webchat": get_astrbot_webchat_path(),  # WebChat 数据
+        "temp": get_astrbot_temp_path(),  # 临时文件
+    }
+
+
+# 备份清单版本号
+BACKUP_MANIFEST_VERSION = "1.1"

astrbot/core/backup/exporter.py

@@ -0,0 +1,477 @@
+"""AstrBot 数据导出器
+
+负责将所有数据导出为 ZIP 备份文件。
+导出格式为 JSON，这是数据库无关的方案，支持未来向 MySQL/PostgreSQL 迁移。
+"""
+
+import hashlib
+import json
+import os
+import zipfile
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import TYPE_CHECKING, Any
+
+from sqlalchemy import select
+
+from astrbot.core import logger
+from astrbot.core.config.default import VERSION
+from astrbot.core.db import BaseDatabase
+from astrbot.core.utils.astrbot_path import (
+    get_astrbot_backups_path,
+    get_astrbot_data_path,
+)
+
+# 从共享常量模块导入
+from .constants import (
+    BACKUP_MANIFEST_VERSION,
+    KB_METADATA_MODELS,
+    MAIN_DB_MODELS,
+    get_backup_directories,
+)
+
+if TYPE_CHECKING:
+    from astrbot.core.knowledge_base.kb_mgr import KnowledgeBaseManager
+
+CMD_CONFIG_FILE_PATH = os.path.join(get_astrbot_data_path(), "cmd_config.json")
+
+
+class AstrBotExporter:
+    """AstrBot 数据导出器
+
+    导出内容：
+    - 主数据库所有表（data/data_v4.db）
+    - 知识库元数据（data/knowledge_base/kb.db）
+    - 每个知识库的向量文档数据
+    - 配置文件（data/cmd_config.json）
+    - 附件文件
+    - 知识库多媒体文件
+    - 插件目录（data/plugins）
+    - 插件数据目录（data/plugin_data）
+    - 配置目录（data/config）
+    - T2I 模板目录（data/t2i_templates）
+    - WebChat 数据目录（data/webchat）
+    - 临时文件目录（data/temp）
+    """
+
+    def __init__(
+        self,
+        main_db: BaseDatabase,
+        kb_manager: "KnowledgeBaseManager | None" = None,
+        config_path: str = CMD_CONFIG_FILE_PATH,
+    ):
+        self.main_db = main_db
+        self.kb_manager = kb_manager
+        self.config_path = config_path
+        self._checksums: dict[str, str] = {}
+
+    async def export_all(
+        self,
+        output_dir: str | None = None,
+        progress_callback: Any | None = None,
+    ) -> str:
+        """导出所有数据到 ZIP 文件
+
+        Args:
+            output_dir: 输出目录
+            progress_callback: 进度回调函数，接收参数 (stage, current, total, message)
+
+        Returns:
+            str: 生成的 ZIP 文件路径
+        """
+        if output_dir is None:
+            output_dir = get_astrbot_backups_path()
+
+        # 确保输出目录存在
+        Path(output_dir).mkdir(parents=True, exist_ok=True)
+
+        timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+        zip_filename = f"astrbot_backup_{timestamp}.zip"
+        zip_path = os.path.join(output_dir, zip_filename)
+
+        logger.info(f"开始导出备份到 {zip_path}")
+
+        try:
+            with zipfile.ZipFile(zip_path, "w", zipfile.ZIP_DEFLATED) as zf:
+                # 1. 导出主数据库
+                if progress_callback:
+                    await progress_callback("main_db", 0, 100, "正在导出主数据库...")
+                main_data = await self._export_main_database()
+                main_db_json = json.dumps(
+                    main_data, ensure_ascii=False, indent=2, default=str
+                )
+                zf.writestr("databases/main_db.json", main_db_json)
+                self._add_checksum("databases/main_db.json", main_db_json)
+                if progress_callback:
+                    await progress_callback("main_db", 100, 100, "主数据库导出完成")
+
+                # 2. 导出知识库数据
+                kb_meta_data: dict[str, Any] = {
+                    "knowledge_bases": [],
+                    "kb_documents": [],
+                    "kb_media": [],
+                }
+                if self.kb_manager:
+                    if progress_callback:
+                        await progress_callback(
+                            "kb_metadata", 0, 100, "正在导出知识库元数据..."
+                        )
+                    kb_meta_data = await self._export_kb_metadata()
+                    kb_meta_json = json.dumps(
+                        kb_meta_data, ensure_ascii=False, indent=2, default=str
+                    )
+                    zf.writestr("databases/kb_metadata.json", kb_meta_json)
+                    self._add_checksum("databases/kb_metadata.json", kb_meta_json)
+                    if progress_callback:
+                        await progress_callback(
+                            "kb_metadata", 100, 100, "知识库元数据导出完成"
+                        )
+
+                    # 导出每个知识库的文档数据
+                    kb_insts = self.kb_manager.kb_insts
+                    total_kbs = len(kb_insts)
+                    for idx, (kb_id, kb_helper) in enumerate(kb_insts.items()):
+                        if progress_callback:
+                            await progress_callback(
+                                "kb_documents",
+                                idx,
+                                total_kbs,
+                                f"正在导出知识库 {kb_helper.kb.kb_name} 的文档数据...",
+                            )
+                        doc_data = await self._export_kb_documents(kb_helper)
+                        doc_json = json.dumps(
+                            doc_data, ensure_ascii=False, indent=2, default=str
+                        )
+                        doc_path = f"databases/kb_{kb_id}/documents.json"
+                        zf.writestr(doc_path, doc_json)
+                        self._add_checksum(doc_path, doc_json)
+
+                        # 导出 FAISS 索引文件
+                        await self._export_faiss_index(zf, kb_helper, kb_id)
+
+                        # 导出知识库多媒体文件
+                        await self._export_kb_media_files(zf, kb_helper, kb_id)
+
+                    if progress_callback:
+                        await progress_callback(
+                            "kb_documents", total_kbs, total_kbs, "知识库文档导出完成"
+                        )
+
+                # 3. 导出配置文件
+                if progress_callback:
+                    await progress_callback("config", 0, 100, "正在导出配置文件...")
+                if os.path.exists(self.config_path):
+                    with open(self.config_path, encoding="utf-8") as f:
+                        config_content = f.read()
+                    zf.writestr("config/cmd_config.json", config_content)
+                    self._add_checksum("config/cmd_config.json", config_content)
+                if progress_callback:
+                    await progress_callback("config", 100, 100, "配置文件导出完成")
+
+                # 4. 导出附件文件
+                if progress_callback:
+                    await progress_callback("attachments", 0, 100, "正在导出附件...")
+                await self._export_attachments(zf, main_data.get("attachments", []))
+                if progress_callback:
+                    await progress_callback("attachments", 100, 100, "附件导出完成")
+
+                # 5. 导出插件和其他目录
+                if progress_callback:
+                    await progress_callback(
+                        "directories", 0, 100, "正在导出插件和数据目录..."
+                    )
+                dir_stats = await self._export_directories(zf)
+                if progress_callback:
+                    await progress_callback("directories", 100, 100, "目录导出完成")
+
+                # 6. 生成 manifest
+                if progress_callback:
+                    await progress_callback("manifest", 0, 100, "正在生成清单...")
+                manifest = self._generate_manifest(main_data, kb_meta_data, dir_stats)
+                manifest_json = json.dumps(manifest, ensure_ascii=False, indent=2)
+                zf.writestr("manifest.json", manifest_json)
+                if progress_callback:
+                    await progress_callback("manifest", 100, 100, "清单生成完成")
+
+            logger.info(f"备份导出完成: {zip_path}")
+            return zip_path
+
+        except Exception as e:
+            logger.error(f"备份导出失败: {e}")
+            # 清理失败的文件
+            if os.path.exists(zip_path):
+                os.remove(zip_path)
+            raise
+
+    async def _export_main_database(self) -> dict[str, list[dict]]:
+        """导出主数据库所有表"""
+        export_data: dict[str, list[dict]] = {}
+
+        async with self.main_db.get_db() as session:
+            for table_name, model_class in MAIN_DB_MODELS.items():
+                try:
+                    result = await session.execute(select(model_class))
+                    records = result.scalars().all()
+                    export_data[table_name] = [
+                        self._model_to_dict(record) for record in records
+                    ]
+                    logger.debug(
+                        f"导出表 {table_name}: {len(export_data[table_name])} 条记录"
+                    )
+                except Exception as e:
+                    logger.warning(f"导出表 {table_name} 失败: {e}")
+                    export_data[table_name] = []
+
+        return export_data
+
+    async def _export_kb_metadata(self) -> dict[str, list[dict]]:
+        """导出知识库元数据库"""
+        if not self.kb_manager:
+            return {"knowledge_bases": [], "kb_documents": [], "kb_media": []}
+
+        export_data: dict[str, list[dict]] = {}
+
+        async with self.kb_manager.kb_db.get_db() as session:
+            for table_name, model_class in KB_METADATA_MODELS.items():
+                try:
+                    result = await session.execute(select(model_class))
+                    records = result.scalars().all()
+                    export_data[table_name] = [
+                        self._model_to_dict(record) for record in records
+                    ]
+                    logger.debug(
+                        f"导出知识库表 {table_name}: {len(export_data[table_name])} 条记录"
+                    )
+                except Exception as e:
+                    logger.warning(f"导出知识库表 {table_name} 失败: {e}")
+                    export_data[table_name] = []
+
+        return export_data
+
+    async def _export_kb_documents(self, kb_helper: Any) -> dict[str, Any]:
+        """导出知识库的文档块数据"""
+        try:
+            from astrbot.core.db.vec_db.faiss_impl.vec_db import FaissVecDB
+
+            vec_db: FaissVecDB = kb_helper.vec_db
+            if not vec_db or not vec_db.document_storage:
+                return {"documents": []}
+
+            # 获取所有文档
+            docs = await vec_db.document_storage.get_documents(
+                metadata_filters={},
+                offset=0,
+                limit=None,  # 获取全部
+            )
+
+            return {"documents": docs}
+        except Exception as e:
+            logger.warning(f"导出知识库文档失败: {e}")
+            return {"documents": []}
+
+    async def _export_faiss_index(
+        self,
+        zf: zipfile.ZipFile,
+        kb_helper: Any,
+        kb_id: str,
+    ) -> None:
+        """导出 FAISS 索引文件"""
+        try:
+            index_path = kb_helper.kb_dir / "index.faiss"
+            if index_path.exists():
+                archive_path = f"databases/kb_{kb_id}/index.faiss"
+                zf.write(str(index_path), archive_path)
+                logger.debug(f"导出 FAISS 索引: {archive_path}")
+        except Exception as e:
+            logger.warning(f"导出 FAISS 索引失败: {e}")
+
+    async def _export_kb_media_files(
+        self, zf: zipfile.ZipFile, kb_helper: Any, kb_id: str
+    ) -> None:
+        """导出知识库的多媒体文件"""
+        try:
+            media_dir = kb_helper.kb_medias_dir
+            if not media_dir.exists():
+                return
+
+            for root, _, files in os.walk(media_dir):
+                for file in files:
+                    file_path = Path(root) / file
+                    # 计算相对路径
+                    rel_path = file_path.relative_to(kb_helper.kb_dir)
+                    archive_path = f"files/kb_media/{kb_id}/{rel_path}"
+                    zf.write(str(file_path), archive_path)
+        except Exception as e:
+            logger.warning(f"导出知识库媒体文件失败: {e}")
+
+    async def _export_directories(
+        self, zf: zipfile.ZipFile
+    ) -> dict[str, dict[str, int]]:
+        """导出插件和其他数据目录
+
+        Returns:
+            dict: 每个目录的统计信息 {dir_name: {"files": count, "size": bytes}}
+        """
+        stats: dict[str, dict[str, int]] = {}
+        backup_directories = get_backup_directories()
+
+        for dir_name, dir_path in backup_directories.items():
+            full_path = Path(dir_path)
+            if not full_path.exists():
+                logger.debug(f"目录不存在，跳过: {full_path}")
+                continue
+
+            file_count = 0
+            total_size = 0
+
+            try:
+                for root, dirs, files in os.walk(full_path):
+                    # 跳过 __pycache__ 目录
+                    dirs[:] = [d for d in dirs if d != "__pycache__"]
+
+                    for file in files:
+                        # 跳过 .pyc 文件
+                        if file.endswith(".pyc"):
+                            continue
+
+                        file_path = Path(root) / file
+                        try:
+                            # 计算相对路径
+                            rel_path = file_path.relative_to(full_path)
+                            archive_path = f"directories/{dir_name}/{rel_path}"
+                            zf.write(str(file_path), archive_path)
+                            file_count += 1
+                            total_size += file_path.stat().st_size
+                        except Exception as e:
+                            logger.warning(f"导出文件 {file_path} 失败: {e}")
+
+                stats[dir_name] = {"files": file_count, "size": total_size}
+                logger.debug(
+                    f"导出目录 {dir_name}: {file_count} 个文件, {total_size} 字节"
+                )
+            except Exception as e:
+                logger.warning(f"导出目录 {dir_path} 失败: {e}")
+                stats[dir_name] = {"files": 0, "size": 0}
+
+        return stats
+
+    async def _export_attachments(
+        self, zf: zipfile.ZipFile, attachments: list[dict]
+    ) -> None:
+        """导出附件文件"""
+        for attachment in attachments:
+            try:
+                file_path = attachment.get("path", "")
+                if file_path and os.path.exists(file_path):
+                    # 使用 attachment_id 作为文件名
+                    attachment_id = attachment.get("attachment_id", "")
+                    ext = os.path.splitext(file_path)[1]
+                    archive_path = f"files/attachments/{attachment_id}{ext}"
+                    zf.write(file_path, archive_path)
+            except Exception as e:
+                logger.warning(f"导出附件失败: {e}")
+
+    def _model_to_dict(self, record: Any) -> dict:
+        """将 SQLModel 实例转换为字典
+
+        这是数据库无关的序列化方式，支持未来迁移到其他数据库。
+        """
+        # 使用 SQLModel 内置的 model_dump 方法（如果可用）
+        if hasattr(record, "model_dump"):
+            data = record.model_dump(mode="python")
+            # 处理 datetime 类型
+            for key, value in data.items():
+                if isinstance(value, datetime):
+                    data[key] = value.isoformat()
+            return data
+
+        # 回退到手动提取
+        data = {}
+        # 使用 inspect 获取表信息
+        from sqlalchemy import inspect as sa_inspect
+
+        mapper = sa_inspect(record.__class__)
+        for column in mapper.columns:
+            value = getattr(record, column.name)
+            # 处理 datetime 类型 - 统一转为 ISO 格式字符串
+            if isinstance(value, datetime):
+                value = value.isoformat()
+            data[column.name] = value
+        return data
+
+    def _add_checksum(self, path: str, content: str | bytes) -> None:
+        """计算并添加文件校验和"""
+        if isinstance(content, str):
+            content = content.encode("utf-8")
+        checksum = hashlib.sha256(content).hexdigest()
+        self._checksums[path] = f"sha256:{checksum}"
+
+    def _generate_manifest(
+        self,
+        main_data: dict[str, list[dict]],
+        kb_meta_data: dict[str, list[dict]],
+        dir_stats: dict[str, dict[str, int]] | None = None,
+    ) -> dict:
+        """生成备份清单"""
+        if dir_stats is None:
+            dir_stats = {}
+        # 收集知识库 ID
+        kb_document_tables = {}
+        if self.kb_manager:
+            for kb_id in self.kb_manager.kb_insts.keys():
+                kb_document_tables[kb_id] = "documents"
+
+        # 收集附件文件列表
+        attachment_files = []
+        for attachment in main_data.get("attachments", []):
+            attachment_id = attachment.get("attachment_id", "")
+            path = attachment.get("path", "")
+            if attachment_id and path:
+                ext = os.path.splitext(path)[1]
+                attachment_files.append(f"{attachment_id}{ext}")
+
+        # 收集知识库媒体文件
+        kb_media_files: dict[str, list[str]] = {}
+        if self.kb_manager:
+            for kb_id, kb_helper in self.kb_manager.kb_insts.items():
+                media_files: list[str] = []
+                media_dir = kb_helper.kb_medias_dir
+                if media_dir.exists():
+                    for root, _, files in os.walk(media_dir):
+                        for file in files:
+                            media_files.append(file)
+                if media_files:
+                    kb_media_files[kb_id] = media_files
+
+        manifest = {
+            "version": BACKUP_MANIFEST_VERSION,
+            "astrbot_version": VERSION,
+            "exported_at": datetime.now(timezone.utc).isoformat(),
+            "origin": "exported",  # 标记备份来源：exported=本实例导出, uploaded=用户上传
+            "schema_version": {
+                "main_db": "v4",
+                "kb_db": "v1",
+            },
+            "tables": {
+                "main_db": list(main_data.keys()),
+                "kb_metadata": list(kb_meta_data.keys()),
+                "kb_documents": kb_document_tables,
+            },
+            "files": {
+                "attachments": attachment_files,
+                "kb_media": kb_media_files,
+            },
+            "directories": list(dir_stats.keys()),
+            "checksums": self._checksums,
+            "statistics": {
+                "main_db": {
+                    table: len(records) for table, records in main_data.items()
+                },
+                "kb_metadata": {
+                    table: len(records) for table, records in kb_meta_data.items()
+                },
+                "directories": dir_stats,
+            },
+        }
+
+        return manifest

astrbot/core/backup/importer.py

@@ -0,0 +1,761 @@
+"""AstrBot 数据导入器
+
+负责从 ZIP 备份文件恢复所有数据。
+导入时进行版本校验：
+- 主版本（前两位）不同时直接拒绝导入
+- 小版本（第三位）不同时提示警告，用户可选择强制导入
+- 版本匹配时也需要用户确认
+"""
+
+import json
+import os
+import shutil
+import zipfile
+from dataclasses import dataclass, field
+from datetime import datetime
+from pathlib import Path
+from typing import TYPE_CHECKING, Any
+
+from sqlalchemy import delete
+
+from astrbot.core import logger
+from astrbot.core.config.default import VERSION
+from astrbot.core.db import BaseDatabase
+from astrbot.core.utils.astrbot_path import (
+    get_astrbot_data_path,
+    get_astrbot_knowledge_base_path,
+)
+from astrbot.core.utils.version_comparator import VersionComparator
+
+# 从共享常量模块导入
+from .constants import (
+    KB_METADATA_MODELS,
+    MAIN_DB_MODELS,
+    get_backup_directories,
+)
+
+if TYPE_CHECKING:
+    from astrbot.core.knowledge_base.kb_mgr import KnowledgeBaseManager
+
+
+def _get_major_version(version_str: str) -> str:
+    """提取版本的主版本部分（前两位）
+
+    Args:
+        version_str: 版本字符串，如 "4.9.1", "4.10.0-beta"
+
+    Returns:
+        主版本字符串，如 "4.9", "4.10"
+    """
+    if not version_str:
+        return "0.0"
+    # 移除 v 前缀和预发布标签
+    version = version_str.lower().replace("v", "").split("-")[0].split("+")[0]
+    parts = [p for p in version.split(".") if p]  # 过滤空字符串
+    if len(parts) >= 2:
+        return f"{parts[0]}.{parts[1]}"
+    elif len(parts) == 1 and parts[0]:
+        return f"{parts[0]}.0"
+    return "0.0"
+
+
+CMD_CONFIG_FILE_PATH = os.path.join(get_astrbot_data_path(), "cmd_config.json")
+KB_PATH = get_astrbot_knowledge_base_path()
+
+
+@dataclass
+class ImportPreCheckResult:
+    """导入预检查结果
+
+    用于在实际导入前检查备份文件的版本兼容性，
+    并返回确认信息让用户决定是否继续导入。
+    """
+
+    # 检查是否通过（文件有效且版本可导入）
+    valid: bool = False
+    # 是否可以导入（版本兼容）
+    can_import: bool = False
+    # 版本状态: match（完全匹配）, minor_diff（小版本差异）, major_diff（主版本不同，拒绝）
+    version_status: str = ""
+    # 备份文件中的 AstrBot 版本
+    backup_version: str = ""
+    # 当前运行的 AstrBot 版本
+    current_version: str = VERSION
+    # 备份创建时间
+    backup_time: str = ""
+    # 确认消息（显示给用户）
+    confirm_message: str = ""
+    # 警告消息列表
+    warnings: list[str] = field(default_factory=list)
+    # 错误消息（如果检查失败）
+    error: str = ""
+    # 备份包含的内容摘要
+    backup_summary: dict = field(default_factory=dict)
+
+    def to_dict(self) -> dict:
+        return {
+            "valid": self.valid,
+            "can_import": self.can_import,
+            "version_status": self.version_status,
+            "backup_version": self.backup_version,
+            "current_version": self.current_version,
+            "backup_time": self.backup_time,
+            "confirm_message": self.confirm_message,
+            "warnings": self.warnings,
+            "error": self.error,
+            "backup_summary": self.backup_summary,
+        }
+
+
+class ImportResult:
+    """导入结果"""
+
+    def __init__(self):
+        self.success = True
+        self.imported_tables: dict[str, int] = {}
+        self.imported_files: dict[str, int] = {}
+        self.imported_directories: dict[str, int] = {}
+        self.warnings: list[str] = []
+        self.errors: list[str] = []
+
+    def add_warning(self, msg: str) -> None:
+        self.warnings.append(msg)
+        logger.warning(msg)
+
+    def add_error(self, msg: str) -> None:
+        self.errors.append(msg)
+        self.success = False
+        logger.error(msg)
+
+    def to_dict(self) -> dict:
+        return {
+            "success": self.success,
+            "imported_tables": self.imported_tables,
+            "imported_files": self.imported_files,
+            "imported_directories": self.imported_directories,
+            "warnings": self.warnings,
+            "errors": self.errors,
+        }
+
+
+class AstrBotImporter:
+    """AstrBot 数据导入器
+
+    导入备份文件中的所有数据，包括：
+    - 主数据库所有表
+    - 知识库元数据和文档
+    - 配置文件
+    - 附件文件
+    - 知识库多媒体文件
+    - 插件目录（data/plugins）
+    - 插件数据目录（data/plugin_data）
+    - 配置目录（data/config）
+    - T2I 模板目录（data/t2i_templates）
+    - WebChat 数据目录（data/webchat）
+    - 临时文件目录（data/temp）
+    """
+
+    def __init__(
+        self,
+        main_db: BaseDatabase,
+        kb_manager: "KnowledgeBaseManager | None" = None,
+        config_path: str = CMD_CONFIG_FILE_PATH,
+        kb_root_dir: str = KB_PATH,
+    ):
+        self.main_db = main_db
+        self.kb_manager = kb_manager
+        self.config_path = config_path
+        self.kb_root_dir = kb_root_dir
+
+    def pre_check(self, zip_path: str) -> ImportPreCheckResult:
+        """预检查备份文件
+
+        在实际导入前检查备份文件的有效性和版本兼容性。
+        返回检查结果供前端显示确认对话框。
+
+        Args:
+            zip_path: ZIP 备份文件路径
+
+        Returns:
+            ImportPreCheckResult: 预检查结果
+        """
+        result = ImportPreCheckResult()
+        result.current_version = VERSION
+
+        if not os.path.exists(zip_path):
+            result.error = f"备份文件不存在: {zip_path}"
+            return result
+
+        try:
+            with zipfile.ZipFile(zip_path, "r") as zf:
+                # 读取 manifest
+                try:
+                    manifest_data = zf.read("manifest.json")
+                    manifest = json.loads(manifest_data)
+                except KeyError:
+                    result.error = "备份文件缺少 manifest.json，不是有效的 AstrBot 备份"
+                    return result
+                except json.JSONDecodeError as e:
+                    result.error = f"manifest.json 格式错误: {e}"
+                    return result
+
+                # 提取基本信息
+                result.backup_version = manifest.get("astrbot_version", "未知")
+                result.backup_time = manifest.get("exported_at", "未知")
+                result.valid = True
+
+                # 构建备份摘要
+                result.backup_summary = {
+                    "tables": list(manifest.get("tables", {}).keys()),
+                    "has_knowledge_bases": manifest.get("has_knowledge_bases", False),
+                    "has_config": manifest.get("has_config", False),
+                    "directories": manifest.get("directories", []),
+                }
+
+                # 检查版本兼容性
+                version_check = self._check_version_compatibility(result.backup_version)
+                result.version_status = version_check["status"]
+                result.can_import = version_check["can_import"]
+
+                # 版本信息由前端根据 version_status 和 i18n 生成显示
+                # 不再将版本消息添加到 warnings 列表中，避免中文硬编码
+                # warnings 列表保留用于其他非版本相关的警告
+
+                return result
+
+        except zipfile.BadZipFile:
+            result.error = "无效的 ZIP 文件"
+            return result
+        except Exception as e:
+            result.error = f"检查备份文件失败: {e}"
+            return result
+
+    def _check_version_compatibility(self, backup_version: str) -> dict:
+        """检查版本兼容性
+
+        规则：
+        - 主版本（前两位，如 4.9）必须一致，否则拒绝
+        - 小版本（第三位，如 4.9.1 vs 4.9.2）不同时，警告但允许导入
+
+        Returns:
+            dict: {status, can_import, message}
+        """
+        if not backup_version:
+            return {
+                "status": "major_diff",
+                "can_import": False,
+                "message": "备份文件缺少版本信息",
+            }
+
+        # 提取主版本（前两位）进行比较
+        backup_major = _get_major_version(backup_version)
+        current_major = _get_major_version(VERSION)
+
+        # 比较主版本
+        if VersionComparator.compare_version(backup_major, current_major) != 0:
+            return {
+                "status": "major_diff",
+                "can_import": False,
+                "message": (
+                    f"主版本不兼容: 备份版本 {backup_version}, 当前版本 {VERSION}。"
+                    f"跨主版本导入可能导致数据损坏，请使用相同主版本的 AstrBot。"
+                ),
+            }
+
+        # 比较完整版本
+        version_cmp = VersionComparator.compare_version(backup_version, VERSION)
+        if version_cmp != 0:
+            return {
+                "status": "minor_diff",
+                "can_import": True,
+                "message": (
+                    f"小版本差异: 备份版本 {backup_version}, 当前版本 {VERSION}。"
+                ),
+            }
+
+        return {
+            "status": "match",
+            "can_import": True,
+            "message": "版本匹配",
+        }
+
+    async def import_all(
+        self,
+        zip_path: str,
+        mode: str = "replace",  # "replace" 清空后导入
+        progress_callback: Any | None = None,
+    ) -> ImportResult:
+        """从 ZIP 文件导入所有数据
+
+        Args:
+            zip_path: ZIP 备份文件路径
+            mode: 导入模式，目前仅支持 "replace"（清空后导入）
+            progress_callback: 进度回调函数，接收参数 (stage, current, total, message)
+
+        Returns:
+            ImportResult: 导入结果
+        """
+        result = ImportResult()
+
+        if not os.path.exists(zip_path):
+            result.add_error(f"备份文件不存在: {zip_path}")
+            return result
+
+        logger.info(f"开始从 {zip_path} 导入备份")
+
+        try:
+            with zipfile.ZipFile(zip_path, "r") as zf:
+                # 1. 读取并验证 manifest
+                if progress_callback:
+                    await progress_callback("validate", 0, 100, "正在验证备份文件...")
+
+                try:
+                    manifest_data = zf.read("manifest.json")
+                    manifest = json.loads(manifest_data)
+                except KeyError:
+                    result.add_error("备份文件缺少 manifest.json")
+                    return result
+                except json.JSONDecodeError as e:
+                    result.add_error(f"manifest.json 格式错误: {e}")
+                    return result
+
+                # 版本校验
+                try:
+                    self._validate_version(manifest)
+                except ValueError as e:
+                    result.add_error(str(e))
+                    return result
+
+                if progress_callback:
+                    await progress_callback("validate", 100, 100, "验证完成")
+
+                # 2. 导入主数据库
+                if progress_callback:
+                    await progress_callback("main_db", 0, 100, "正在导入主数据库...")
+
+                try:
+                    main_data_content = zf.read("databases/main_db.json")
+                    main_data = json.loads(main_data_content)
+
+                    if mode == "replace":
+                        await self._clear_main_db()
+
+                    imported = await self._import_main_database(main_data)
+                    result.imported_tables.update(imported)
+                except Exception as e:
+                    result.add_error(f"导入主数据库失败: {e}")
+                    return result
+
+                if progress_callback:
+                    await progress_callback("main_db", 100, 100, "主数据库导入完成")
+
+                # 3. 导入知识库
+                if self.kb_manager and "databases/kb_metadata.json" in zf.namelist():
+                    if progress_callback:
+                        await progress_callback("kb", 0, 100, "正在导入知识库...")
+
+                    try:
+                        kb_meta_content = zf.read("databases/kb_metadata.json")
+                        kb_meta_data = json.loads(kb_meta_content)
+
+                        if mode == "replace":
+                            await self._clear_kb_data()
+
+                        await self._import_knowledge_bases(zf, kb_meta_data, result)
+                    except Exception as e:
+                        result.add_warning(f"导入知识库失败: {e}")
+
+                    if progress_callback:
+                        await progress_callback("kb", 100, 100, "知识库导入完成")
+
+                # 4. 导入配置文件
+                if progress_callback:
+                    await progress_callback("config", 0, 100, "正在导入配置文件...")
+
+                if "config/cmd_config.json" in zf.namelist():
+                    try:
+                        config_content = zf.read("config/cmd_config.json")
+                        # 备份现有配置
+                        if os.path.exists(self.config_path):
+                            backup_path = f"{self.config_path}.bak"
+                            shutil.copy2(self.config_path, backup_path)
+
+                        with open(self.config_path, "wb") as f:
+                            f.write(config_content)
+                        result.imported_files["config"] = 1
+                    except Exception as e:
+                        result.add_warning(f"导入配置文件失败: {e}")
+
+                if progress_callback:
+                    await progress_callback("config", 100, 100, "配置文件导入完成")
+
+                # 5. 导入附件文件
+                if progress_callback:
+                    await progress_callback("attachments", 0, 100, "正在导入附件...")
+
+                attachment_count = await self._import_attachments(
+                    zf, main_data.get("attachments", [])
+                )
+                result.imported_files["attachments"] = attachment_count
+
+                if progress_callback:
+                    await progress_callback("attachments", 100, 100, "附件导入完成")
+
+                # 6. 导入插件和其他目录
+                if progress_callback:
+                    await progress_callback(
+                        "directories", 0, 100, "正在导入插件和数据目录..."
+                    )
+
+                dir_stats = await self._import_directories(zf, manifest, result)
+                result.imported_directories = dir_stats
+
+                if progress_callback:
+                    await progress_callback("directories", 100, 100, "目录导入完成")
+
+            logger.info(f"备份导入完成: {result.to_dict()}")
+            return result
+
+        except zipfile.BadZipFile:
+            result.add_error("无效的 ZIP 文件")
+            return result
+        except Exception as e:
+            result.add_error(f"导入失败: {e}")
+            return result
+
+    def _validate_version(self, manifest: dict) -> None:
+        """验证版本兼容性 - 仅允许相同主版本导入
+
+        注意：此方法仅在 import_all 中调用，用于双重校验。
+        前端应先调用 pre_check 获取详细的版本信息并让用户确认。
+        """
+        backup_version = manifest.get("astrbot_version")
+        if not backup_version:
+            raise ValueError("备份文件缺少版本信息")
+
+        # 使用新的版本兼容性检查
+        version_check = self._check_version_compatibility(backup_version)
+
+        if version_check["status"] == "major_diff":
+            raise ValueError(version_check["message"])
+
+        # minor_diff 和 match 都允许导入
+        if version_check["status"] == "minor_diff":
+            logger.warning(f"版本差异警告: {version_check['message']}")
+
+    async def _clear_main_db(self) -> None:
+        """清空主数据库所有表"""
+        async with self.main_db.get_db() as session:
+            async with session.begin():
+                for table_name, model_class in MAIN_DB_MODELS.items():
+                    try:
+                        await session.execute(delete(model_class))
+                        logger.debug(f"已清空表 {table_name}")
+                    except Exception as e:
+                        logger.warning(f"清空表 {table_name} 失败: {e}")
+
+    async def _clear_kb_data(self) -> None:
+        """清空知识库数据"""
+        if not self.kb_manager:
+            return
+
+        # 清空知识库元数据表
+        async with self.kb_manager.kb_db.get_db() as session:
+            async with session.begin():
+                for table_name, model_class in KB_METADATA_MODELS.items():
+                    try:
+                        await session.execute(delete(model_class))
+                        logger.debug(f"已清空知识库表 {table_name}")
+                    except Exception as e:
+                        logger.warning(f"清空知识库表 {table_name} 失败: {e}")
+
+        # 删除知识库文件目录
+        for kb_id in list(self.kb_manager.kb_insts.keys()):
+            try:
+                kb_helper = self.kb_manager.kb_insts[kb_id]
+                await kb_helper.terminate()
+                if kb_helper.kb_dir.exists():
+                    shutil.rmtree(kb_helper.kb_dir)
+            except Exception as e:
+                logger.warning(f"清理知识库 {kb_id} 失败: {e}")
+
+        self.kb_manager.kb_insts.clear()
+
+    async def _import_main_database(
+        self, data: dict[str, list[dict]]
+    ) -> dict[str, int]:
+        """导入主数据库数据"""
+        imported: dict[str, int] = {}
+
+        async with self.main_db.get_db() as session:
+            async with session.begin():
+                for table_name, rows in data.items():
+                    model_class = MAIN_DB_MODELS.get(table_name)
+                    if not model_class:
+                        logger.warning(f"未知的表: {table_name}")
+                        continue
+
+                    count = 0
+                    for row in rows:
+                        try:
+                            # 转换 datetime 字符串为 datetime 对象
+                            row = self._convert_datetime_fields(row, model_class)
+                            obj = model_class(**row)
+                            session.add(obj)
+                            count += 1
+                        except Exception as e:
+                            logger.warning(f"导入记录到 {table_name} 失败: {e}")
+
+                    imported[table_name] = count
+                    logger.debug(f"导入表 {table_name}: {count} 条记录")
+
+        return imported
+
+    async def _import_knowledge_bases(
+        self,
+        zf: zipfile.ZipFile,
+        kb_meta_data: dict[str, list[dict]],
+        result: ImportResult,
+    ) -> None:
+        """导入知识库数据"""
+        if not self.kb_manager:
+            return
+
+        # 1. 导入知识库元数据
+        async with self.kb_manager.kb_db.get_db() as session:
+            async with session.begin():
+                for table_name, rows in kb_meta_data.items():
+                    model_class = KB_METADATA_MODELS.get(table_name)
+                    if not model_class:
+                        continue
+
+                    count = 0
+                    for row in rows:
+                        try:
+                            row = self._convert_datetime_fields(row, model_class)
+                            obj = model_class(**row)
+                            session.add(obj)
+                            count += 1
+                        except Exception as e:
+                            logger.warning(f"导入知识库记录到 {table_name} 失败: {e}")
+
+                    result.imported_tables[f"kb_{table_name}"] = count
+
+        # 2. 导入每个知识库的文档和文件
+        for kb_data in kb_meta_data.get("knowledge_bases", []):
+            kb_id = kb_data.get("kb_id")
+            if not kb_id:
+                continue
+
+            # 创建知识库目录
+            kb_dir = Path(self.kb_root_dir) / kb_id
+            kb_dir.mkdir(parents=True, exist_ok=True)
+
+            # 导入文档数据
+            doc_path = f"databases/kb_{kb_id}/documents.json"
+            if doc_path in zf.namelist():
+                try:
+                    doc_content = zf.read(doc_path)
+                    doc_data = json.loads(doc_content)
+
+                    # 导入到文档存储数据库
+                    await self._import_kb_documents(kb_id, doc_data)
+                except Exception as e:
+                    result.add_warning(f"导入知识库 {kb_id} 的文档失败: {e}")
+
+            # 导入 FAISS 索引
+            faiss_path = f"databases/kb_{kb_id}/index.faiss"
+            if faiss_path in zf.namelist():
+                try:
+                    target_path = kb_dir / "index.faiss"
+                    with zf.open(faiss_path) as src, open(target_path, "wb") as dst:
+                        dst.write(src.read())
+                except Exception as e:
+                    result.add_warning(f"导入知识库 {kb_id} 的 FAISS 索引失败: {e}")
+
+            # 导入媒体文件
+            media_prefix = f"files/kb_media/{kb_id}/"
+            for name in zf.namelist():
+                if name.startswith(media_prefix):
+                    try:
+                        rel_path = name[len(media_prefix) :]
+                        target_path = kb_dir / rel_path
+                        target_path.parent.mkdir(parents=True, exist_ok=True)
+                        with zf.open(name) as src, open(target_path, "wb") as dst:
+                            dst.write(src.read())
+                    except Exception as e:
+                        result.add_warning(f"导入媒体文件 {name} 失败: {e}")
+
+        # 3. 重新加载知识库实例
+        await self.kb_manager.load_kbs()
+
+    async def _import_kb_documents(self, kb_id: str, doc_data: dict) -> None:
+        """导入知识库文档到向量数据库"""
+        from astrbot.core.db.vec_db.faiss_impl.document_storage import DocumentStorage
+
+        kb_dir = Path(self.kb_root_dir) / kb_id
+        doc_db_path = kb_dir / "doc.db"
+
+        # 初始化文档存储
+        doc_storage = DocumentStorage(str(doc_db_path))
+        await doc_storage.initialize()
+
+        try:
+            documents = doc_data.get("documents", [])
+            for doc in documents:
+                try:
+                    await doc_storage.insert_document(
+                        doc_id=doc.get("doc_id", ""),
+                        text=doc.get("text", ""),
+                        metadata=json.loads(doc.get("metadata", "{}")),
+                    )
+                except Exception as e:
+                    logger.warning(f"导入文档块失败: {e}")
+        finally:
+            await doc_storage.close()
+
+    async def _import_attachments(
+        self,
+        zf: zipfile.ZipFile,
+        attachments: list[dict],
+    ) -> int:
+        """导入附件文件"""
+        count = 0
+
+        attachments_dir = Path(self.config_path).parent / "attachments"
+        attachments_dir.mkdir(parents=True, exist_ok=True)
+
+        attachment_prefix = "files/attachments/"
+        for name in zf.namelist():
+            if name.startswith(attachment_prefix) and name != attachment_prefix:
+                try:
+                    # 从附件记录中找到原始路径
+                    attachment_id = os.path.splitext(os.path.basename(name))[0]
+                    original_path = None
+                    for att in attachments:
+                        if att.get("attachment_id") == attachment_id:
+                            original_path = att.get("path")
+                            break
+
+                    if original_path:
+                        target_path = Path(original_path)
+                    else:
+                        target_path = attachments_dir / os.path.basename(name)
+
+                    target_path.parent.mkdir(parents=True, exist_ok=True)
+                    with zf.open(name) as src, open(target_path, "wb") as dst:
+                        dst.write(src.read())
+                    count += 1
+                except Exception as e:
+                    logger.warning(f"导入附件 {name} 失败: {e}")
+
+        return count
+
+    async def _import_directories(
+        self,
+        zf: zipfile.ZipFile,
+        manifest: dict,
+        result: ImportResult,
+    ) -> dict[str, int]:
+        """导入插件和其他数据目录
+
+        Args:
+            zf: ZIP 文件对象
+            manifest: 备份清单
+            result: 导入结果对象
+
+        Returns:
+            dict: 每个目录导入的文件数量
+        """
+        dir_stats: dict[str, int] = {}
+
+        # 检查备份版本是否支持目录备份（需要版本 >= 1.1）
+        backup_version = manifest.get("version", "1.0")
+        if VersionComparator.compare_version(backup_version, "1.1") < 0:
+            logger.info("备份版本不支持目录备份，跳过目录导入")
+            return dir_stats
+
+        backed_up_dirs = manifest.get("directories", [])
+        backup_directories = get_backup_directories()
+
+        for dir_name in backed_up_dirs:
+            if dir_name not in backup_directories:
+                result.add_warning(f"未知的目录类型: {dir_name}")
+                continue
+
+            target_dir = Path(backup_directories[dir_name])
+            archive_prefix = f"directories/{dir_name}/"
+
+            file_count = 0
+
+            try:
+                # 获取该目录下的所有文件
+                dir_files = [
+                    name
+                    for name in zf.namelist()
+                    if name.startswith(archive_prefix) and name != archive_prefix
+                ]
+
+                if not dir_files:
+                    continue
+
+                # 备份现有目录（如果存在）
+                if target_dir.exists():
+                    backup_path = Path(f"{target_dir}.bak")
+                    if backup_path.exists():
+                        shutil.rmtree(backup_path)
+                    shutil.move(str(target_dir), str(backup_path))
+                    logger.debug(f"已备份现有目录 {target_dir} 到 {backup_path}")
+
+                # 创建目标目录
+                target_dir.mkdir(parents=True, exist_ok=True)
+
+                # 解压文件
+                for name in dir_files:
+                    try:
+                        # 计算相对路径
+                        rel_path = name[len(archive_prefix) :]
+                        if not rel_path:  # 跳过目录条目
+                            continue
+
+                        target_path = target_dir / rel_path
+                        target_path.parent.mkdir(parents=True, exist_ok=True)
+
+                        with zf.open(name) as src, open(target_path, "wb") as dst:
+                            dst.write(src.read())
+                        file_count += 1
+                    except Exception as e:
+                        result.add_warning(f"导入文件 {name} 失败: {e}")
+
+                dir_stats[dir_name] = file_count
+                logger.debug(f"导入目录 {dir_name}: {file_count} 个文件")
+
+            except Exception as e:
+                result.add_warning(f"导入目录 {dir_name} 失败: {e}")
+                dir_stats[dir_name] = 0
+
+        return dir_stats
+
+    def _convert_datetime_fields(self, row: dict, model_class: type) -> dict:
+        """转换 datetime 字符串字段为 datetime 对象"""
+        result = row.copy()
+
+        # 获取模型的 datetime 字段
+        from sqlalchemy import inspect as sa_inspect
+
+        try:
+            mapper = sa_inspect(model_class)
+            for column in mapper.columns:
+                if column.name in result and result[column.name] is not None:
+                    # 检查是否是 datetime 类型的列
+                    from sqlalchemy import DateTime
+
+                    if isinstance(column.type, DateTime):
+                        value = result[column.name]
+                        if isinstance(value, str):
+                            # 解析 ISO 格式的日期时间字符串
+                            result[column.name] = datetime.fromisoformat(value)
+        except Exception:
+            pass
+
+        return result

astrbot/core/config/astrbot_config.py

@@ -80,6 +80,8 @@ class AstrBotConfig(dict):
                 if v["type"] == "object":
                     conf[k] = {}
                     _parse_schema(v["items"], conf[k])
+                elif v["type"] == "template_list":
+                    conf[k] = default
                 else:
                     conf[k] = default
 

astrbot/core/config/default.py

@@ -5,7 +5,7 @@ from typing import Any, TypedDict
 
 from astrbot.core.utils.astrbot_path import get_astrbot_data_path
 
-VERSION = "4.10.0"
+VERSION = "4.11.0"
 DB_PATH = os.path.join(get_astrbot_data_path(), "data_v4.db")
 
 WEBHOOK_SUPPORTED_PLATFORMS = [
@@ -83,6 +83,16 @@ DEFAULT_CONFIG = {
         "default_personality": "default",
         "persona_pool": ["*"],
         "prompt_prefix": "{{prompt}}",
+        "context_limit_reached_strategy": "truncate_by_turns",  # or llm_compress
+        "llm_compress_instruction": (
+            "Based on our full conversation history, produce a concise summary of key takeaways and/or project progress.\n"
+            "1. Systematically cover all core topics discussed and the final conclusion/outcome for each; clearly highlight the latest primary focus.\n"
+            "2. If any tools were used, summarize tool usage (total call count) and extract the most valuable insights from tool outputs.\n"
+            "3. If there was an initial user goal, state it first and describe the current progress/status.\n"
+            "4. Write the summary in the user's language.\n"
+        ),
+        "llm_compress_keep_recent": 4,
+        "llm_compress_provider_id": "",
         "max_context_length": -1,
         "dequeue_context_length": 1,
         "streaming_response": False,
@@ -179,6 +189,7 @@ class ChatProviderTemplate(TypedDict):
     model: str
     modalities: list
     custom_extra_body: dict[str, Any]
+    max_context_tokens: int
 
 
 CHAT_PROVIDER_TEMPLATE = {
@@ -187,6 +198,7 @@ CHAT_PROVIDER_TEMPLATE = {
     "model": "",
     "modalities": [],
     "custom_extra_body": {},
+    "max_context_tokens": 0,
 }
 
 """
@@ -227,7 +239,7 @@ CONFIG_METADATA_2 = {
                         "callback_server_host": "0.0.0.0",
                         "port": 6196,
                     },
-                    "OneBot v11": {
+                    "OneBot v11 (QQ 个人号等)": {
                         "id": "default",
                         "type": "aiocqhttp",
                         "enable": False,
@@ -235,16 +247,6 @@ CONFIG_METADATA_2 = {
                         "ws_reverse_port": 6199,
                         "ws_reverse_token": "",
                     },
-                    "WeChatPadPro": {
-                        "id": "wechatpadpro",
-                        "type": "wechatpadpro",
-                        "enable": False,
-                        "admin_key": "stay33",
-                        "host": "这里填写你的局域网IP或者公网服务器IP",
-                        "port": 8059,
-                        "wpp_active_message_poll": False,
-                        "wpp_active_message_poll_interval": 3,
-                    },
                     "微信公众平台": {
                         "id": "weixin_official_account",
                         "type": "weixin_official_account",
@@ -374,6 +376,16 @@ CONFIG_METADATA_2 = {
                         "satori_heartbeat_interval": 10,
                         "satori_reconnect_delay": 5,
                     },
+                    "WeChatPadPro": {
+                        "id": "wechatpadpro",
+                        "type": "wechatpadpro",
+                        "enable": False,
+                        "admin_key": "stay33",
+                        "host": "这里填写你的局域网IP或者公网服务器IP",
+                        "port": 8059,
+                        "wpp_active_message_poll": False,
+                        "wpp_active_message_poll_interval": 3,
+                    },
                     # "WebChat": {
                     #     "id": "webchat",
                     #     "type": "webchat",
@@ -905,6 +917,7 @@ CONFIG_METADATA_2 = {
                         "key": [],
                         "api_base": "https://api.anthropic.com/v1",
                         "timeout": 120,
+                        "anth_thinking_config": {"budget": 0},
                     },
                     "Moonshot": {
                         "id": "moonshot",
@@ -920,7 +933,7 @@ CONFIG_METADATA_2 = {
                     "xAI": {
                         "id": "xai",
                         "provider": "xai",
-                        "type": "openai_chat_completion",
+                        "type": "xai_chat_completion",
                         "provider_type": "chat_completion",
                         "enable": True,
                         "key": [],
@@ -1286,7 +1299,7 @@ CONFIG_METADATA_2 = {
                         "minimax-is-timber-weight": False,
                         "minimax-voice-id": "female-shaonv",
                         "minimax-timber-weight": '[\n    {\n        "voice_id": "Chinese (Mandarin)_Warm_Girl",\n        "weight": 25\n    },\n    {\n        "voice_id": "Chinese (Mandarin)_BashfulGirl",\n        "weight": 50\n    }\n]',
-                        "minimax-voice-emotion": "neutral",
+                        "minimax-voice-emotion": "auto",
                         "minimax-voice-latex": False,
                         "minimax-voice-english-normalization": False,
                         "timeout": 20,
@@ -1450,7 +1463,32 @@ CONFIG_METADATA_2 = {
                         "description": "自定义请求体参数",
                         "type": "dict",
                         "items": {},
-                        "hint": "此处添加的键值对将被合并到发送给 API 的 extra_body 中。值可以是字符串、数字或布尔值。",
+                        "hint": "用于在请求时添加额外的参数，如 temperature、top_p、max_tokens 等。",
+                        "template_schema": {
+                            "temperature": {
+                                "name": "Temperature",
+                                "description": "温度参数",
+                                "hint": "控制输出的随机性，范围通常为 0-2。值越高越随机。",
+                                "type": "float",
+                                "default": 0.6,
+                                "slider": {"min": 0, "max": 2, "step": 0.1},
+                            },
+                            "top_p": {
+                                "name": "Top-p",
+                                "description": "Top-p 采样",
+                                "hint": "核采样参数，范围通常为 0-1。控制模型考虑的概率质量。",
+                                "type": "float",
+                                "default": 1.0,
+                                "slider": {"min": 0, "max": 1, "step": 0.01},
+                            },
+                            "max_tokens": {
+                                "name": "Max Tokens",
+                                "description": "最大令牌数",
+                                "hint": "生成的最大令牌数。",
+                                "type": "int",
+                                "default": 8192,
+                            },
+                        },
                     },
                     "provider": {
                         "type": "string",
@@ -1787,6 +1825,17 @@ CONFIG_METADATA_2 = {
                             },
                         },
                     },
+                    "anth_thinking_config": {
+                        "description": "Thinking Config",
+                        "type": "object",
+                        "items": {
+                            "budget": {
+                                "description": "Thinking Budget",
+                                "type": "int",
+                                "hint": "Anthropic thinking.budget_tokens param. Must >= 1024. See: https://platform.claude.com/docs/en/build-with-claude/extended-thinking",
+                            },
+                        },
+                    },
                     "minimax-group-id": {
                         "type": "string",
                         "description": "用户组",
@@ -1858,15 +1907,18 @@ CONFIG_METADATA_2 = {
                     "minimax-voice-emotion": {
                         "type": "string",
                         "description": "情绪",
-                        "hint": "控制合成语音的情绪",
+                        "hint": "控制合成语音的情绪。当为 auto 时，将根据文本内容自动选择情绪。",
                         "options": [
+                            "auto",
                             "happy",
                             "sad",
                             "angry",
                             "fearful",
                             "disgusted",
                             "surprised",
-                            "neutral",
+                            "calm",
+                            "fluent",
+                            "whisper",
                         ],
                     },
                     "minimax-voice-latex": {
@@ -1993,6 +2045,11 @@ CONFIG_METADATA_2 = {
                         "type": "string",
                         "hint": "模型名称，如 gpt-4o-mini, deepseek-chat。",
                     },
+                    "max_context_tokens": {
+                        "description": "模型上下文窗口大小",
+                        "type": "int",
+                        "hint": "模型最大上下文 Token 大小。如果为 0，则会自动从模型元数据填充（如有），也可手动修改。",
+                    },
                     "dify_api_key": {
                         "description": "API Key",
                         "type": "string",
@@ -2500,6 +2557,66 @@ CONFIG_METADATA_3 = {
             #         "provider_settings.enable": True,
             #     },
             # },
+            "truncate_and_compress": {
+                "description": "上下文管理策略",
+                "type": "object",
+                "items": {
+                    "provider_settings.max_context_length": {
+                        "description": "最多携带对话轮数",
+                        "type": "int",
+                        "hint": "超出这个数量时丢弃最旧的部分，一轮聊天记为 1 条，-1 为不限制",
+                        "condition": {
+                            "provider_settings.agent_runner_type": "local",
+                        },
+                    },
+                    "provider_settings.dequeue_context_length": {
+                        "description": "丢弃对话轮数",
+                        "type": "int",
+                        "hint": "超出最多携带对话轮数时, 一次丢弃的聊天轮数",
+                        "condition": {
+                            "provider_settings.agent_runner_type": "local",
+                        },
+                    },
+                    "provider_settings.context_limit_reached_strategy": {
+                        "description": "超出模型上下文窗口时的处理方式",
+                        "type": "string",
+                        "options": ["truncate_by_turns", "llm_compress"],
+                        "labels": ["按对话轮数截断", "由 LLM 压缩上下文"],
+                        "condition": {
+                            "provider_settings.agent_runner_type": "local",
+                        },
+                        "hint": "",
+                    },
+                    "provider_settings.llm_compress_instruction": {
+                        "description": "上下文压缩提示词",
+                        "type": "text",
+                        "hint": "如果为空则使用默认提示词。",
+                        "condition": {
+                            "provider_settings.context_limit_reached_strategy": "llm_compress",
+                            "provider_settings.agent_runner_type": "local",
+                        },
+                    },
+                    "provider_settings.llm_compress_keep_recent": {
+                        "description": "压缩时保留最近对话轮数",
+                        "type": "int",
+                        "hint": "始终保留的最近 N 轮对话。",
+                        "condition": {
+                            "provider_settings.context_limit_reached_strategy": "llm_compress",
+                            "provider_settings.agent_runner_type": "local",
+                        },
+                    },
+                    "provider_settings.llm_compress_provider_id": {
+                        "description": "用于上下文压缩的模型提供商 ID",
+                        "type": "string",
+                        "_special": "select_provider",
+                        "hint": "留空时将降级为“按对话轮数截断”的策略。",
+                        "condition": {
+                            "provider_settings.context_limit_reached_strategy": "llm_compress",
+                            "provider_settings.agent_runner_type": "local",
+                        },
+                    },
+                },
+            },
             "others": {
                 "description": "其他配置",
                 "type": "object",
@@ -2564,22 +2681,6 @@ CONFIG_METADATA_3 = {
                             "provider_settings.streaming_response": True,
                         },
                     },
-                    "provider_settings.max_context_length": {
-                        "description": "最多携带对话轮数",
-                        "type": "int",
-                        "hint": "超出这个数量时丢弃最旧的部分，一轮聊天记为 1 条，-1 为不限制",
-                        "condition": {
-                            "provider_settings.agent_runner_type": "local",
-                        },
-                    },
-                    "provider_settings.dequeue_context_length": {
-                        "description": "丢弃对话轮数",
-                        "type": "int",
-                        "hint": "超出最多携带对话轮数时, 一次丢弃的聊天轮数",
-                        "condition": {
-                            "provider_settings.agent_runner_type": "local",
-                        },
-                    },
                     "provider_settings.wake_prefix": {
                         "description": "LLM 聊天额外唤醒前缀 ",
                         "type": "string",
@@ -3049,4 +3150,5 @@ DEFAULT_VALUE_MAP = {
     "text": "",
     "list": [],
     "object": {},
+    "template_list": [],
 }

astrbot/core/conversation_mgr.py

@@ -69,6 +69,7 @@ class ConversationManager:
             persona_id=conv_v2.persona_id,
             created_at=created_at,
             updated_at=updated_at,
+            token_usage=conv_v2.token_usage,
         )
 
     async def new_conversation(
@@ -256,6 +257,7 @@ class ConversationManager:
         history: list[dict] | None = None,
         title: str | None = None,
         persona_id: str | None = None,
+        token_usage: int | None = None,
     ) -> None:
         """更新会话的对话.
 
@@ -263,6 +265,7 @@ class ConversationManager:
             unified_msg_origin (str): 统一的消息来源字符串。格式为 platform_name:message_type:session_id
             conversation_id (str): 对话 ID, 是 uuid 格式的字符串
             history (List[Dict]): 对话历史记录, 是一个字典列表, 每个字典包含 role 和 content 字段
+            token_usage (int | None): token 使用量。None 表示不更新
 
         """
         if not conversation_id:
@@ -274,6 +277,7 @@ class ConversationManager:
                 title=title,
                 persona_id=persona_id,
                 content=history,
+                token_usage=token_usage,
             )
 
     async def update_conversation_title(

astrbot/core/core_lifecycle.py

@@ -90,6 +90,7 @@ class AstrBotCoreLifecycle:
 
         # 初始化 UMOP 配置路由器
         self.umop_config_router = UmopConfigRouter(sp=sp)
+        await self.umop_config_router.initialize()
 
         # 初始化 AstrBot 配置管理器
         self.astrbot_config_mgr = AstrBotConfigManager(

astrbot/core/db/__init__.py

@@ -152,6 +152,7 @@ class BaseDatabase(abc.ABC):
         title: str | None = None,
         persona_id: str | None = None,
         content: list[dict] | None = None,
+        token_usage: int | None = None,
     ) -> None:
         """Update a conversation's history."""
         ...

astrbot/core/db/migration/migra_token_usage.py

@@ -0,0 +1,61 @@
+"""Migration script to add token_usage column to conversations table.
+
+This migration adds the token_usage field to track token consumption for each conversation.
+
+Changes:
+- Adds token_usage column to conversations table (default: 0)
+"""
+
+from sqlalchemy import text
+
+from astrbot.api import logger, sp
+from astrbot.core.db import BaseDatabase
+
+
+async def migrate_token_usage(db_helper: BaseDatabase):
+    """Add token_usage column to conversations table.
+
+    This migration adds a new column to track token consumption in conversations.
+    """
+    # 检查是否已经完成迁移
+    migration_done = await db_helper.get_preference(
+        "global", "global", "migration_done_token_usage_1"
+    )
+    if migration_done:
+        return
+
+    logger.info("开始执行数据库迁移（添加 conversations.token_usage 列）...")
+
+    # 这里只适配了 SQLite。因为截止至这一版本，AstrBot 仅支持 SQLite。
+
+    try:
+        async with db_helper.get_db() as session:
+            # 检查列是否已存在
+            result = await session.execute(text("PRAGMA table_info(conversations)"))
+            columns = result.fetchall()
+            column_names = [col[1] for col in columns]
+
+            if "token_usage" in column_names:
+                logger.info("token_usage 列已存在，跳过迁移")
+                await sp.put_async(
+                    "global", "global", "migration_done_token_usage_1", True
+                )
+                return
+
+            # 添加 token_usage 列
+            await session.execute(
+                text(
+                    "ALTER TABLE conversations ADD COLUMN token_usage INTEGER NOT NULL DEFAULT 0"
+                )
+            )
+            await session.commit()
+
+            logger.info("token_usage 列添加成功")
+
+        # 标记迁移完成
+        await sp.put_async("global", "global", "migration_done_token_usage_1", True)
+        logger.info("token_usage 迁移完成")
+
+    except Exception as e:
+        logger.error(f"迁移过程中发生错误: {e}", exc_info=True)
+        raise

astrbot/core/db/po.py

@@ -54,6 +54,11 @@ class ConversationV2(SQLModel, table=True):
     )
     title: str | None = Field(default=None, max_length=255)
     persona_id: str | None = Field(default=None)
+    token_usage: int = Field(default=0, nullable=False)
+    """content is a list of OpenAI-formated messages in list[dict] format.
+    token_usage is the total token value of the messages.
+    when 0, will use estimated token counter.
+    """
 
     __table_args__ = (
         UniqueConstraint(
@@ -313,6 +318,8 @@ class Conversation:
     persona_id: str | None = ""
     created_at: int = 0
     updated_at: int = 0
+    token_usage: int = 0
+    """对话的总 token 数量。AstrBot 会保留最近一次 LLM 请求返回的总 token 数，方便统计。token_usage 可能为 0，表示未知。"""
 
 
 class Personality(TypedDict):

astrbot/core/db/sqlite.py

@@ -241,7 +241,9 @@ class SQLiteDatabase(BaseDatabase):
                 session.add(new_conversation)
                 return new_conversation
 
-    async def update_conversation(self, cid, title=None, persona_id=None, content=None):
+    async def update_conversation(
+        self, cid, title=None, persona_id=None, content=None, token_usage=None
+    ):
         async with self.get_db() as session:
             session: AsyncSession
             async with session.begin():
@@ -255,6 +257,8 @@ class SQLiteDatabase(BaseDatabase):
                     values["persona_id"] = persona_id
                 if content is not None:
                     values["content"] = content
+                if token_usage is not None:
+                    values["token_usage"] = token_usage
                 if not values:
                     return None
                 query = query.values(**values)

astrbot/core/knowledge_base/chunking/recursive.py

@@ -149,8 +149,16 @@ class RecursiveCharacterChunker(BaseChunker):
             分割后的文本块列表
 
         """
-        chunk_size = chunk_size or self.chunk_size
-        overlap = overlap or self.chunk_overlap
+        if chunk_size is None:
+            chunk_size = self.chunk_size
+        if overlap is None:
+            overlap = self.chunk_overlap
+        if chunk_size <= 0:
+            raise ValueError("chunk_size must be greater than 0")
+        if overlap < 0:
+            raise ValueError("chunk_overlap must be non-negative")
+        if overlap >= chunk_size:
+            raise ValueError("chunk_overlap must be less than chunk_size")
         result = []
         for i in range(0, len(text), chunk_size - overlap):
             end = min(i + chunk_size, len(text))

astrbot/core/log.py

@@ -58,7 +58,7 @@ def is_plugin_path(pathname):
         return False
 
     norm_path = os.path.normpath(pathname)
-    return ("data/plugins" in norm_path) or ("packages/" in norm_path)
+    return ("data/plugins" in norm_path) or ("astrbot/builtin_stars/" in norm_path)
 
 
 def get_short_level_name(level_name):

astrbot/core/pipeline/process_stage/method/agent_request.py

@@ -38,7 +38,7 @@ class AgentRequestSubStage(Stage):
             )
             return
 
-        if not SessionServiceManager.should_process_llm_request(event):
+        if not await SessionServiceManager.should_process_llm_request(event):
             logger.debug(
                 f"The session {event.unified_msg_origin} has disabled AI capability, skipping processing."
             )

astrbot/core/pipeline/process_stage/method/agent_sub_stages/internal.py

@@ -1,11 +1,12 @@
 """本地 Agent 模式的 LLM 调用 Stage"""
 
 import asyncio
-import copy
 import json
 from collections.abc import AsyncGenerator
 
 from astrbot.core import logger
+from astrbot.core.agent.message import Message
+from astrbot.core.agent.response import AgentStats
 from astrbot.core.agent.tool import ToolSet
 from astrbot.core.astr_agent_context import AstrAgentContext
 from astrbot.core.conversation_mgr import Conversation
@@ -23,6 +24,7 @@ from astrbot.core.provider.entities import (
 )
 from astrbot.core.star.star_handler import EventType, star_map
 from astrbot.core.utils.file_extract import extract_file_moonshotai
+from astrbot.core.utils.llm_metadata import LLM_METADATAS
 from astrbot.core.utils.metrics import Metric
 from astrbot.core.utils.session_lock import session_lock_manager
 
@@ -40,11 +42,6 @@ class InternalAgentSubStage(Stage):
         self.ctx = ctx
         conf = ctx.astrbot_config
         settings = conf["provider_settings"]
-        self.max_context_length = settings["max_context_length"]  # int
-        self.dequeue_context_length: int = min(
-            max(1, settings["dequeue_context_length"]),
-            self.max_context_length - 1,
-        )
         self.streaming_response: bool = settings["streaming_response"]
         self.unsupported_streaming_strategy: str = settings[
             "unsupported_streaming_strategy"
@@ -64,6 +61,25 @@ class InternalAgentSubStage(Stage):
             "moonshotai_api_key", ""
         )
 
+        # 上下文管理相关
+        self.context_limit_reached_strategy: str = settings.get(
+            "context_limit_reached_strategy", "truncate_by_turns"
+        )
+        self.llm_compress_instruction: str = settings.get(
+            "llm_compress_instruction", ""
+        )
+        self.llm_compress_keep_recent: int = settings.get("llm_compress_keep_recent", 4)
+        self.llm_compress_provider_id: str = settings.get(
+            "llm_compress_provider_id", ""
+        )
+        self.max_context_length = settings["max_context_length"]  # int
+        self.dequeue_context_length: int = min(
+            max(1, settings["dequeue_context_length"]),
+            self.max_context_length - 1,
+        )
+        if self.dequeue_context_length <= 0:
+            self.dequeue_context_length = 1
+
         self.conv_manager = ctx.plugin_manager.context.conversation_manager
 
     def _select_provider(self, event: AstrMessageEvent):
@@ -166,34 +182,6 @@ class InternalAgentSubStage(Stage):
                 },
             )
 
-    def _truncate_contexts(
-        self,
-        contexts: list[dict],
-    ) -> list[dict]:
-        """截断上下文列表，确保不超过最大长度"""
-        if self.max_context_length == -1:
-            return contexts
-
-        if len(contexts) // 2 <= self.max_context_length:
-            return contexts
-
-        truncated_contexts = contexts[
-            -(self.max_context_length - self.dequeue_context_length + 1) * 2 :
-        ]
-        # 找到第一个role 为 user 的索引，确保上下文格式正确
-        index = next(
-            (
-                i
-                for i, item in enumerate(truncated_contexts)
-                if item.get("role") == "user"
-            ),
-            None,
-        )
-        if index is not None and index > 0:
-            truncated_contexts = truncated_contexts[index:]
-
-        return truncated_contexts
-
     def _modalities_fix(
         self,
         provider: Provider,
@@ -294,6 +282,8 @@ class InternalAgentSubStage(Stage):
         event: AstrMessageEvent,
         req: ProviderRequest,
         llm_response: LLMResponse | None,
+        all_messages: list[Message],
+        runner_stats: AgentStats | None,
     ):
         if (
             not req
@@ -307,222 +297,255 @@ class InternalAgentSubStage(Stage):
             logger.debug("LLM 响应为空，不保存记录。")
             return
 
-        if req.contexts is None:
-            req.contexts = []
+        # using agent context messages to save to history
+        message_to_save = []
+        for message in all_messages:
+            if message.role == "system":
+                # we do not save system messages to history
+                continue
+            if message.role in ["assistant", "user"] and getattr(
+                message, "_no_save", None
+            ):
+                # we do not save user and assistant messages that are marked as _no_save
+                continue
+            message_to_save.append(message.model_dump())
+
+        # get token usage from agent runner stats
+        token_usage = None
+        if runner_stats:
+            token_usage = runner_stats.token_usage.total
 
-        # 历史上下文
-        messages = copy.deepcopy(req.contexts)
-        # 这一轮对话请求的用户输入
-        messages.append(await req.assemble_context())
-        # 这一轮对话的 LLM 响应
-        if req.tool_calls_result:
-            if not isinstance(req.tool_calls_result, list):
-                messages.extend(req.tool_calls_result.to_openai_messages())
-            elif isinstance(req.tool_calls_result, list):
-                for tcr in req.tool_calls_result:
-                    messages.extend(tcr.to_openai_messages())
-        messages.append(
-            {
-                "role": "assistant",
-                "content": llm_response.completion_text or "*No response*",
-            }
-        )
-        messages = list(filter(lambda item: "_no_save" not in item, messages))
         await self.conv_manager.update_conversation(
             event.unified_msg_origin,
             req.conversation.cid,
-            history=messages,
+            history=message_to_save,
+            token_usage=token_usage,
         )
 
-    def _fix_messages(self, messages: list[dict]) -> list[dict]:
-        """验证并且修复上下文"""
-        fixed_messages = []
-        for message in messages:
-            if message.get("role") == "tool":
-                # tool block 前面必须要有 user 和 assistant block
-                if len(fixed_messages) < 2:
-                    # 这种情况可能是上下文被截断导致的
-                    # 我们直接将之前的上下文都清空
-                    fixed_messages = []
-                else:
-                    fixed_messages.append(message)
-            else:
-                fixed_messages.append(message)
-        return fixed_messages
+    def _get_compress_provider(self) -> Provider | None:
+        if not self.llm_compress_provider_id:
+            return None
+        if self.context_limit_reached_strategy != "llm_compress":
+            return None
+        provider = self.ctx.plugin_manager.context.get_provider_by_id(
+            self.llm_compress_provider_id,
+        )
+        if provider is None:
+            logger.warning(
+                f"未找到指定的上下文压缩模型 {self.llm_compress_provider_id}，将跳过压缩。",
+            )
+            return None
+        if not isinstance(provider, Provider):
+            logger.warning(
+                f"指定的上下文压缩模型 {self.llm_compress_provider_id} 不是对话模型，将跳过压缩。"
+            )
+            return None
+        return provider
 
     async def process(
         self, event: AstrMessageEvent, provider_wake_prefix: str
     ) -> AsyncGenerator[None, None]:
         req: ProviderRequest | None = None
 
-        provider = self._select_provider(event)
-        if provider is None:
-            return
-        if not isinstance(provider, Provider):
-            logger.error(f"选择的提供商类型无效({type(provider)})，跳过 LLM 请求处理。")
-            return
-
-        streaming_response = self.streaming_response
-        if (enable_streaming := event.get_extra("enable_streaming")) is not None:
-            streaming_response = bool(enable_streaming)
-
-        logger.debug("ready to request llm provider")
-        async with session_lock_manager.acquire_lock(event.unified_msg_origin):
-            logger.debug("acquired session lock for llm request")
-            if event.get_extra("provider_request"):
-                req = event.get_extra("provider_request")
-                assert isinstance(req, ProviderRequest), (
-                    "provider_request 必须是 ProviderRequest 类型。"
+        try:
+            provider = self._select_provider(event)
+            if provider is None:
+                return
+            if not isinstance(provider, Provider):
+                logger.error(
+                    f"选择的提供商类型无效({type(provider)})，跳过 LLM 请求处理。"
                 )
+                return
 
-                if req.conversation:
-                    req.contexts = json.loads(req.conversation.history)
+            streaming_response = self.streaming_response
+            if (enable_streaming := event.get_extra("enable_streaming")) is not None:
+                streaming_response = bool(enable_streaming)
 
-            else:
-                req = ProviderRequest()
-                req.prompt = ""
-                req.image_urls = []
-                if sel_model := event.get_extra("selected_model"):
-                    req.model = sel_model
-                if provider_wake_prefix and not event.message_str.startswith(
-                    provider_wake_prefix
-                ):
+            logger.debug("ready to request llm provider")
+
+            # 通知等待调用 LLM（在获取锁之前）
+            await call_event_hook(event, EventType.OnWaitingLLMRequestEvent)
+
+            async with session_lock_manager.acquire_lock(event.unified_msg_origin):
+                logger.debug("acquired session lock for llm request")
+                if event.get_extra("provider_request"):
+                    req = event.get_extra("provider_request")
+                    assert isinstance(req, ProviderRequest), (
+                        "provider_request 必须是 ProviderRequest 类型。"
+                    )
+
+                    if req.conversation:
+                        req.contexts = json.loads(req.conversation.history)
+
+                else:
+                    req = ProviderRequest()
+                    req.prompt = ""
+                    req.image_urls = []
+                    if sel_model := event.get_extra("selected_model"):
+                        req.model = sel_model
+                    if provider_wake_prefix and not event.message_str.startswith(
+                        provider_wake_prefix
+                    ):
+                        return
+
+                    req.prompt = event.message_str[len(provider_wake_prefix) :]
+                    # func_tool selection 现在已经转移到 astrbot/builtin_stars/astrbot 插件中进行选择。
+                    # req.func_tool = self.ctx.plugin_manager.context.get_llm_tool_manager()
+                    for comp in event.message_obj.message:
+                        if isinstance(comp, Image):
+                            image_path = await comp.convert_to_file_path()
+                            req.image_urls.append(image_path)
+
+                    conversation = await self._get_session_conv(event)
+                    req.conversation = conversation
+                    req.contexts = json.loads(conversation.history)
+
+                    event.set_extra("provider_request", req)
+
+                # fix contexts json str
+                if isinstance(req.contexts, str):
+                    req.contexts = json.loads(req.contexts)
+
+                # apply file extract
+                if self.file_extract_enabled:
+                    try:
+                        await self._apply_file_extract(event, req)
+                    except Exception as e:
+                        logger.error(f"Error occurred while applying file extract: {e}")
+
+                if not req.prompt and not req.image_urls:
                     return
 
-                req.prompt = event.message_str[len(provider_wake_prefix) :]
-                # func_tool selection 现在已经转移到 packages/astrbot 插件中进行选择。
-                # req.func_tool = self.ctx.plugin_manager.context.get_llm_tool_manager()
-                for comp in event.message_obj.message:
-                    if isinstance(comp, Image):
-                        image_path = await comp.convert_to_file_path()
-                        req.image_urls.append(image_path)
+                # call event hook
+                if await call_event_hook(event, EventType.OnLLMRequestEvent, req):
+                    return
 
-                conversation = await self._get_session_conv(event)
-                req.conversation = conversation
-                req.contexts = json.loads(conversation.history)
+                # apply knowledge base feature
+                await self._apply_kb(event, req)
 
-                event.set_extra("provider_request", req)
+                # truncate contexts to fit max length
+                # NOW moved to ContextManager inside ToolLoopAgentRunner
+                # if req.contexts:
+                #     req.contexts = self._truncate_contexts(req.contexts)
+                #     self._fix_messages(req.contexts)
 
-            # fix contexts json str
-            if isinstance(req.contexts, str):
-                req.contexts = json.loads(req.contexts)
+                # session_id
+                if not req.session_id:
+                    req.session_id = event.unified_msg_origin
 
-            # apply file extract
-            if self.file_extract_enabled:
-                try:
-                    await self._apply_file_extract(event, req)
-                except Exception as e:
-                    logger.error(f"Error occurred while applying file extract: {e}")
+                # check provider modalities, if provider does not support image/tool_use, clear them in request.
+                self._modalities_fix(provider, req)
 
-            if not req.prompt and not req.image_urls:
-                return
+                # filter tools, only keep tools from this pipeline's selected plugins
+                self._plugin_tool_fix(event, req)
 
-            # call event hook
-            if await call_event_hook(event, EventType.OnLLMRequestEvent, req):
-                return
-
-            # apply knowledge base feature
-            await self._apply_kb(event, req)
-
-            # truncate contexts to fit max length
-            if req.contexts:
-                req.contexts = self._truncate_contexts(req.contexts)
-                self._fix_messages(req.contexts)
-
-            # session_id
-            if not req.session_id:
-                req.session_id = event.unified_msg_origin
-
-            # check provider modalities, if provider does not support image/tool_use, clear them in request.
-            self._modalities_fix(provider, req)
-
-            # filter tools, only keep tools from this pipeline's selected plugins
-            self._plugin_tool_fix(event, req)
-
-            stream_to_general = (
-                self.unsupported_streaming_strategy == "turn_off"
-                and not event.platform_meta.support_streaming_message
-            )
-            # 备份 req.contexts
-            backup_contexts = copy.deepcopy(req.contexts)
-
-            # run agent
-            agent_runner = AgentRunner()
-            logger.debug(
-                f"handle provider[id: {provider.provider_config['id']}] request: {req}",
-            )
-            astr_agent_ctx = AstrAgentContext(
-                context=self.ctx.plugin_manager.context,
-                event=event,
-            )
-            await agent_runner.reset(
-                provider=provider,
-                request=req,
-                run_context=AgentContextWrapper(
-                    context=astr_agent_ctx,
-                    tool_call_timeout=self.tool_call_timeout,
-                ),
-                tool_executor=FunctionToolExecutor(),
-                agent_hooks=MAIN_AGENT_HOOKS,
-                streaming=streaming_response,
-            )
-
-            if streaming_response and not stream_to_general:
-                # 流式响应
-                event.set_result(
-                    MessageEventResult()
-                    .set_result_content_type(ResultContentType.STREAMING_RESULT)
-                    .set_async_stream(
-                        run_agent(
-                            agent_runner,
-                            self.max_step,
-                            self.show_tool_use,
-                            show_reasoning=self.show_reasoning,
-                        ),
-                    ),
+                stream_to_general = (
+                    self.unsupported_streaming_strategy == "turn_off"
+                    and not event.platform_meta.support_streaming_message
                 )
-                yield
-                if agent_runner.done():
-                    if final_llm_resp := agent_runner.get_final_llm_resp():
-                        if final_llm_resp.completion_text:
-                            chain = (
-                                MessageChain()
-                                .message(final_llm_resp.completion_text)
-                                .chain
-                            )
-                        elif final_llm_resp.result_chain:
-                            chain = final_llm_resp.result_chain.chain
-                        else:
-                            chain = MessageChain().chain
-                        event.set_result(
-                            MessageEventResult(
-                                chain=chain,
-                                result_content_type=ResultContentType.STREAMING_FINISH,
+
+                # run agent
+                agent_runner = AgentRunner()
+                logger.debug(
+                    f"handle provider[id: {provider.provider_config['id']}] request: {req}",
+                )
+                astr_agent_ctx = AstrAgentContext(
+                    context=self.ctx.plugin_manager.context,
+                    event=event,
+                )
+
+                # inject model context length limit
+                if provider.provider_config.get("max_context_tokens", 0) <= 0:
+                    model = provider.get_model()
+                    if model_info := LLM_METADATAS.get(model):
+                        provider.provider_config["max_context_tokens"] = model_info[
+                            "limit"
+                        ]["context"]
+
+                await agent_runner.reset(
+                    provider=provider,
+                    request=req,
+                    run_context=AgentContextWrapper(
+                        context=astr_agent_ctx,
+                        tool_call_timeout=self.tool_call_timeout,
+                    ),
+                    tool_executor=FunctionToolExecutor(),
+                    agent_hooks=MAIN_AGENT_HOOKS,
+                    streaming=streaming_response,
+                    llm_compress_instruction=self.llm_compress_instruction,
+                    llm_compress_keep_recent=self.llm_compress_keep_recent,
+                    llm_compress_provider=self._get_compress_provider(),
+                    truncate_turns=self.dequeue_context_length,
+                    enforce_max_turns=self.max_context_length,
+                )
+
+                if streaming_response and not stream_to_general:
+                    # 流式响应
+                    event.set_result(
+                        MessageEventResult()
+                        .set_result_content_type(ResultContentType.STREAMING_RESULT)
+                        .set_async_stream(
+                            run_agent(
+                                agent_runner,
+                                self.max_step,
+                                self.show_tool_use,
+                                show_reasoning=self.show_reasoning,
                             ),
-                        )
-            else:
-                async for _ in run_agent(
-                    agent_runner,
-                    self.max_step,
-                    self.show_tool_use,
-                    stream_to_general,
-                    show_reasoning=self.show_reasoning,
-                ):
+                        ),
+                    )
                     yield
+                    if agent_runner.done():
+                        if final_llm_resp := agent_runner.get_final_llm_resp():
+                            if final_llm_resp.completion_text:
+                                chain = (
+                                    MessageChain()
+                                    .message(final_llm_resp.completion_text)
+                                    .chain
+                                )
+                            elif final_llm_resp.result_chain:
+                                chain = final_llm_resp.result_chain.chain
+                            else:
+                                chain = MessageChain().chain
+                            event.set_result(
+                                MessageEventResult(
+                                    chain=chain,
+                                    result_content_type=ResultContentType.STREAMING_FINISH,
+                                ),
+                            )
+                else:
+                    async for _ in run_agent(
+                        agent_runner,
+                        self.max_step,
+                        self.show_tool_use,
+                        stream_to_general,
+                        show_reasoning=self.show_reasoning,
+                    ):
+                        yield
 
-            # 恢复备份的 contexts
-            req.contexts = backup_contexts
+                await self._save_to_history(
+                    event,
+                    req,
+                    agent_runner.get_final_llm_resp(),
+                    agent_runner.run_context.messages,
+                    agent_runner.stats,
+                )
 
-            await self._save_to_history(event, req, agent_runner.get_final_llm_resp())
+            # 异步处理 WebChat 特殊情况
+            if event.get_platform_name() == "webchat":
+                asyncio.create_task(self._handle_webchat(event, req, provider))
 
-        # 异步处理 WebChat 特殊情况
-        if event.get_platform_name() == "webchat":
-            asyncio.create_task(self._handle_webchat(event, req, provider))
+            asyncio.create_task(
+                Metric.upload(
+                    llm_tick=1,
+                    model_name=agent_runner.provider.get_model(),
+                    provider_type=agent_runner.provider.meta().type,
+                ),
+            )
 
-        asyncio.create_task(
-            Metric.upload(
-                llm_tick=1,
-                model_name=agent_runner.provider.get_model(),
-                provider_type=agent_runner.provider.meta().type,
-            ),
-        )
+        except Exception as e:
+            logger.error(f"Error occurred while processing agent: {e}")
+            await event.send(
+                MessageChain().message(
+                    f"Error occurred while processing agent request: {e}"
+                )
+            )

astrbot/core/pipeline/result_decorate/stage.py

@@ -98,6 +98,9 @@ class ResultDecorateStage(Stage):
                     self.content_safe_check_stage = stage_cls()
                     await self.content_safe_check_stage.initialize(ctx)
 
+        provider_cfg = ctx.astrbot_config.get("provider_settings", {})
+        self.show_reasoning = provider_cfg.get("display_reasoning_text", False)
+
     def _split_text_by_words(self, text: str) -> list[str]:
         """使用分段词列表分段文本"""
         if not self.split_words_pattern:
@@ -254,70 +257,75 @@ class ResultDecorateStage(Stage):
                 event.unified_msg_origin,
             )
 
-            if (
-                self.ctx.astrbot_config["provider_tts_settings"]["enable"]
+            should_tts = (
+                bool(self.ctx.astrbot_config["provider_tts_settings"]["enable"])
                 and result.is_llm_result()
-                and SessionServiceManager.should_process_tts_request(event)
-            ):
-                should_tts = self.tts_trigger_probability >= 1.0 or (
-                    self.tts_trigger_probability > 0.0
-                    and random.random() <= self.tts_trigger_probability
+                and await SessionServiceManager.should_process_tts_request(event)
+                and random.random() <= self.tts_trigger_probability
+                and tts_provider
+            )
+            if should_tts and not tts_provider:
+                logger.warning(
+                    f"会话 {event.unified_msg_origin} 未配置文本转语音模型。",
                 )
 
-                if not should_tts:
-                    logger.debug("跳过 TTS：触发概率未命中。")
-                elif not tts_provider:
-                    logger.warning(
-                        f"会话 {event.unified_msg_origin} 未配置文本转语音模型。",
-                    )
-                else:
-                    new_chain = []
-                    for comp in result.chain:
-                        if isinstance(comp, Plain) and len(comp.text) > 1:
-                            try:
-                                logger.info(f"TTS 请求: {comp.text}")
-                                audio_path = await tts_provider.get_audio(comp.text)
-                                logger.info(f"TTS 结果: {audio_path}")
-                                if not audio_path:
-                                    logger.error(
-                                        f"由于 TTS 音频文件未找到，消息段转语音失败: {comp.text}",
-                                    )
-                                    new_chain.append(comp)
-                                    continue
+            if (
+                not should_tts
+                and self.show_reasoning
+                and event.get_extra("_llm_reasoning_content")
+            ):
+                # inject reasoning content to chain
+                reasoning_content = event.get_extra("_llm_reasoning_content")
+                result.chain.insert(0, Plain(f"🤔 思考: {reasoning_content}\n"))
 
-                                use_file_service = self.ctx.astrbot_config[
-                                    "provider_tts_settings"
-                                ]["use_file_service"]
-                                callback_api_base = self.ctx.astrbot_config[
-                                    "callback_api_base"
-                                ]
-                                dual_output = self.ctx.astrbot_config[
-                                    "provider_tts_settings"
-                                ]["dual_output"]
-
-                                url = None
-                                if use_file_service and callback_api_base:
-                                    token = await file_token_service.register_file(
-                                        audio_path,
-                                    )
-                                    url = f"{callback_api_base}/api/file/{token}"
-                                    logger.debug(f"已注册：{url}")
-
-                                new_chain.append(
-                                    Record(
-                                        file=url or audio_path,
-                                        url=url or audio_path,
-                                    ),
+            if should_tts and tts_provider:
+                new_chain = []
+                for comp in result.chain:
+                    if isinstance(comp, Plain) and len(comp.text) > 1:
+                        try:
+                            logger.info(f"TTS 请求: {comp.text}")
+                            audio_path = await tts_provider.get_audio(comp.text)
+                            logger.info(f"TTS 结果: {audio_path}")
+                            if not audio_path:
+                                logger.error(
+                                    f"由于 TTS 音频文件未找到，消息段转语音失败: {comp.text}",
                                 )
-                                if dual_output:
-                                    new_chain.append(comp)
-                            except Exception:
-                                logger.error(traceback.format_exc())
-                                logger.error("TTS 失败，使用文本发送。")
                                 new_chain.append(comp)
-                        else:
+                                continue
+
+                            use_file_service = self.ctx.astrbot_config[
+                                "provider_tts_settings"
+                            ]["use_file_service"]
+                            callback_api_base = self.ctx.astrbot_config[
+                                "callback_api_base"
+                            ]
+                            dual_output = self.ctx.astrbot_config[
+                                "provider_tts_settings"
+                            ]["dual_output"]
+
+                            url = None
+                            if use_file_service and callback_api_base:
+                                token = await file_token_service.register_file(
+                                    audio_path,
+                                )
+                                url = f"{callback_api_base}/api/file/{token}"
+                                logger.debug(f"已注册：{url}")
+
+                            new_chain.append(
+                                Record(
+                                    file=url or audio_path,
+                                    url=url or audio_path,
+                                ),
+                            )
+                            if dual_output:
+                                new_chain.append(comp)
+                        except Exception:
+                            logger.error(traceback.format_exc())
+                            logger.error("TTS 失败，使用文本发送。")
                             new_chain.append(comp)
-                    result.chain = new_chain
+                    else:
+                        new_chain.append(comp)
+                result.chain = new_chain
 
             # 文本转图片
             elif (

astrbot/core/pipeline/session_status_check/stage.py

@@ -21,7 +21,7 @@ class SessionStatusCheckStage(Stage):
         event: AstrMessageEvent,
     ) -> None | AsyncGenerator[None, None]:
         # 检查会话是否整体启用
-        if not SessionServiceManager.is_session_enabled(event.unified_msg_origin):
+        if not await SessionServiceManager.is_session_enabled(event.unified_msg_origin):
             logger.debug(f"会话 {event.unified_msg_origin} 已被关闭，已终止事件传播。")
 
             # workaround for #2309

astrbot/core/pipeline/waking_check/stage.py

@@ -1,9 +1,10 @@
-from collections.abc import AsyncGenerator
+from collections.abc import AsyncGenerator, Callable
 
 from astrbot import logger
 from astrbot.core.message.components import At, AtAll, Reply
 from astrbot.core.message.message_event_result import MessageChain, MessageEventResult
 from astrbot.core.platform.astr_message_event import AstrMessageEvent
+from astrbot.core.platform.message_type import MessageType
 from astrbot.core.star.filter.command_group import CommandGroupFilter
 from astrbot.core.star.filter.permission import PermissionTypeFilter
 from astrbot.core.star.session_plugin_manager import SessionPluginManager
@@ -13,6 +14,23 @@ from astrbot.core.star.star_handler import EventType, star_handlers_registry
 from ..context import PipelineContext
 from ..stage import Stage, register_stage
 
+UNIQUE_SESSION_ID_BUILDERS: dict[str, Callable[[AstrMessageEvent], str | None]] = {
+    "aiocqhttp": lambda e: f"{e.get_sender_id()}_{e.get_group_id()}",
+    "slack": lambda e: f"{e.get_sender_id()}_{e.get_group_id()}",
+    "dingtalk": lambda e: e.get_sender_id(),
+    "qq_official": lambda e: e.get_sender_id(),
+    "qq_official_webhook": lambda e: e.get_sender_id(),
+    "lark": lambda e: f"{e.get_sender_id()}%{e.get_group_id()}",
+    "misskey": lambda e: f"{e.get_session_id()}_{e.get_sender_id()}",
+    "wechatpadpro": lambda e: f"{e.get_group_id()}#{e.get_sender_id()}",
+}
+
+
+def build_unique_session_id(event: AstrMessageEvent) -> str | None:
+    platform = event.get_platform_name()
+    builder = UNIQUE_SESSION_ID_BUILDERS.get(platform)
+    return builder(event) if builder else None
+
 
 @register_stage
 class WakingCheckStage(Stage):
@@ -53,18 +71,27 @@ class WakingCheckStage(Stage):
         self.disable_builtin_commands = self.ctx.astrbot_config.get(
             "disable_builtin_commands", False
         )
+        platform_settings = self.ctx.astrbot_config.get("platform_settings", {})
+        self.unique_session = platform_settings.get("unique_session", False)
 
     async def process(
         self,
         event: AstrMessageEvent,
     ) -> None | AsyncGenerator[None, None]:
+        # apply unique session
+        if self.unique_session and event.message_obj.type == MessageType.GROUP_MESSAGE:
+            sid = build_unique_session_id(event)
+            if sid:
+                event.session_id = sid
+
+        # ignore bot self message
         if (
             self.ignore_bot_self_message
             and event.get_self_id() == event.get_sender_id()
         ):
-            # 忽略机器人自己发送的消息
             event.stop_event()
             return
+
         # 设置 sender 身份
         event.message_str = event.message_str.strip()
         for admin_id in self.ctx.astrbot_config["admins_id"]:
@@ -136,7 +163,8 @@ class WakingCheckStage(Stage):
         ):
             if (
                 self.disable_builtin_commands
-                and handler.handler_module_path == "packages.builtin_commands.main"
+                and handler.handler_module_path
+                == "astrbot.builtin_stars.builtin_commands.main"
             ):
                 logger.debug("skipping builtin command")
                 continue
@@ -199,7 +227,7 @@ class WakingCheckStage(Stage):
             event._extras.pop("parsed_params", None)
 
         # 根据会话配置过滤插件处理器
-        activated_handlers = SessionPluginManager.filter_handlers_by_session(
+        activated_handlers = await SessionPluginManager.filter_handlers_by_session(
             event,
             activated_handlers,
         )

astrbot/core/platform/sources/aiocqhttp/aiocqhttp_platform_adapter.py

@@ -41,7 +41,6 @@ class AiocqhttpAdapter(Platform):
         super().__init__(platform_config, event_queue)
 
         self.settings = platform_settings
-        self.unique_session = platform_settings["unique_session"]
         self.host = platform_config["ws_reverse_host"]
         self.port = platform_config["ws_reverse_port"]
 
@@ -136,14 +135,11 @@ class AiocqhttpAdapter(Platform):
             abm.group_id = str(event.group_id)
         else:
             abm.type = MessageType.FRIEND_MESSAGE
-        if self.unique_session and abm.type == MessageType.GROUP_MESSAGE:
-            abm.session_id = str(abm.sender.user_id) + "_" + str(event.group_id)
-        else:
-            abm.session_id = (
-                str(event.group_id)
-                if abm.type == MessageType.GROUP_MESSAGE
-                else abm.sender.user_id
-            )
+        abm.session_id = (
+            str(event.group_id)
+            if abm.type == MessageType.GROUP_MESSAGE
+            else abm.sender.user_id
+        )
         abm.message_str = ""
         abm.message = []
         abm.timestamp = int(time.time())
@@ -164,16 +160,11 @@ class AiocqhttpAdapter(Platform):
             abm.type = MessageType.GROUP_MESSAGE
         else:
             abm.type = MessageType.FRIEND_MESSAGE
-        if self.unique_session and abm.type == MessageType.GROUP_MESSAGE:
-            abm.session_id = (
-                str(abm.sender.user_id) + "_" + str(event.group_id)
-            )  # 也保留群组 id
-        else:
-            abm.session_id = (
-                str(event.group_id)
-                if abm.type == MessageType.GROUP_MESSAGE
-                else abm.sender.user_id
-            )
+        abm.session_id = (
+            str(event.group_id)
+            if abm.type == MessageType.GROUP_MESSAGE
+            else abm.sender.user_id
+        )
         abm.message_str = ""
         abm.message = []
         abm.raw_message = event
@@ -210,16 +201,11 @@ class AiocqhttpAdapter(Platform):
             abm.group.group_name = event.get("group_name", "N/A")
         elif event["message_type"] == "private":
             abm.type = MessageType.FRIEND_MESSAGE
-        if self.unique_session and abm.type == MessageType.GROUP_MESSAGE:
-            abm.session_id = (
-                abm.sender.user_id + "_" + str(event.group_id)
-            )  # 也保留群组 id
-        else:
-            abm.session_id = (
-                str(event.group_id)
-                if abm.type == MessageType.GROUP_MESSAGE
-                else abm.sender.user_id
-            )
+        abm.session_id = (
+            str(event.group_id)
+            if abm.type == MessageType.GROUP_MESSAGE
+            else abm.sender.user_id
+        )
 
         abm.message_id = str(event.message_id)
         abm.message = []
@@ -385,10 +371,25 @@ class AiocqhttpAdapter(Platform):
                         logger.error(f"获取 @ 用户信息失败: {e}，此消息段将被忽略。")
 
                 message_str += "".join(at_parts)
+            elif t == "markdown":
+                text = m["data"].get("markdown") or m["data"].get("content", "")
+                abm.message.append(Plain(text=text))
+                message_str += text
             else:
                 for m in m_group:
-                    a = ComponentTypes[t](**m["data"])
-                    abm.message.append(a)
+                    try:
+                        if t not in ComponentTypes:
+                            logger.warning(
+                                f"不支持的消息段类型，已忽略: {t}, data={m['data']}"
+                            )
+                            continue
+                        a = ComponentTypes[t](**m["data"])
+                        abm.message.append(a)
+                    except Exception as e:
+                        logger.exception(
+                            f"消息段解析失败: type={t}, data={m['data']}. {e}"
+                        )
+                        continue
 
         abm.timestamp = int(time.time())
         abm.message_str = message_str

astrbot/core/platform/sources/dingtalk/dingtalk_adapter.py

@@ -50,8 +50,6 @@ class DingtalkPlatformAdapter(Platform):
     ) -> None:
         super().__init__(platform_config, event_queue)
 
-        self.unique_session = platform_settings["unique_session"]
-
         self.client_id = platform_config["client_id"]
         self.client_secret = platform_config["client_secret"]
 
@@ -129,10 +127,7 @@ class DingtalkPlatformAdapter(Platform):
                     if id := self._id_to_sid(user.dingtalk_id):
                         abm.message.append(At(qq=id))
             abm.group_id = message.conversation_id
-            if self.unique_session:
-                abm.session_id = abm.sender.user_id
-            else:
-                abm.session_id = abm.group_id
+            abm.session_id = abm.group_id
         else:
             abm.session_id = abm.sender.user_id
 

astrbot/core/platform/sources/dingtalk/dingtalk_event.py

@@ -25,6 +25,20 @@ class DingtalkMessageEvent(AstrMessageEvent):
         client: dingtalk_stream.ChatbotHandler,
         message: MessageChain,
     ):
+        icm = cast(dingtalk_stream.ChatbotMessage, self.message_obj.raw_message)
+        ats = []
+        # fixes: #4218
+        # 钉钉 at 机器人需要使用 sender_staff_id 而不是 sender_id
+        for i in message.chain:
+            if isinstance(i, Comp.At):
+                print(i.qq, icm.sender_id, icm.sender_staff_id)
+                if str(i.qq) in str(icm.sender_id or ""):
+                    # 适配器会将开头的 $:LWCP_v1:$ 去掉，因此我们用 in 判断
+                    ats.append(f"@{icm.sender_staff_id}")
+                else:
+                    ats.append(f"@{i.qq}")
+        at_str = " ".join(ats)
+
         for segment in message.chain:
             if isinstance(segment, Comp.Plain):
                 segment.text = segment.text.strip()
@@ -32,7 +46,7 @@ class DingtalkMessageEvent(AstrMessageEvent):
                     None,
                     client.reply_markdown,
                     segment.text,
-                    segment.text,
+                    f"{at_str} {segment.text}".strip(),
                     cast(dingtalk_stream.ChatbotMessage, self.message_obj.raw_message),
                 )
             elif isinstance(segment, Comp.Image):

astrbot/core/platform/sources/lark/lark_adapter.py

@@ -44,8 +44,6 @@ class LarkPlatformAdapter(Platform):
     ) -> None:
         super().__init__(platform_config, event_queue)
 
-        self.unique_session = platform_settings["unique_session"]
-
         self.appid = platform_config["app_id"]
         self.appsecret = platform_config["app_secret"]
         self.domain = platform_config.get("domain", lark.FEISHU_DOMAIN)
@@ -317,14 +315,8 @@ class LarkPlatformAdapter(Platform):
             user_id=event.event.sender.sender_id.open_id,
             nickname=event.event.sender.sender_id.open_id[:8],
         )
-        # 独立会话
-        if not self.unique_session:
-            if abm.type == MessageType.GROUP_MESSAGE:
-                abm.session_id = abm.group_id
-            else:
-                abm.session_id = abm.sender.user_id
-        elif abm.type == MessageType.GROUP_MESSAGE:
-            abm.session_id = f"{abm.sender.user_id}%{abm.group_id}"  # 也保留群组id
+        if abm.type == MessageType.GROUP_MESSAGE:
+            abm.session_id = abm.group_id
         else:
             abm.session_id = abm.sender.user_id
 

astrbot/core/platform/sources/misskey/misskey_adapter.py

@@ -91,8 +91,6 @@ class MisskeyPlatformAdapter(Platform):
         except Exception:
             self.max_download_bytes = None
 
-        self.unique_session = platform_settings["unique_session"]
-
         self.api: MisskeyAPI | None = None
         self._running = False
         self.client_self_id = ""
@@ -641,7 +639,6 @@ class MisskeyPlatformAdapter(Platform):
             sender_info,
             self.client_self_id,
             is_chat=False,
-            unique_session=self.unique_session,
         )
         cache_user_info(
             self._user_cache,
@@ -690,7 +687,6 @@ class MisskeyPlatformAdapter(Platform):
             sender_info,
             self.client_self_id,
             is_chat=True,
-            unique_session=self.unique_session,
         )
         cache_user_info(
             self._user_cache,
@@ -720,7 +716,6 @@ class MisskeyPlatformAdapter(Platform):
             self.client_self_id,
             is_chat=False,
             room_id=room_id,
-            unique_session=self.unique_session,
         )
 
         cache_user_info(

astrbot/core/platform/sources/misskey/misskey_utils.py

@@ -338,7 +338,6 @@ def create_base_message(
     client_self_id: str,
     is_chat: bool = False,
     room_id: str | None = None,
-    unique_session: bool = False,
 ) -> AstrBotMessage:
     """创建基础消息对象"""
     message = AstrBotMessage()
@@ -353,8 +352,6 @@ def create_base_message(
     if room_id:
         session_prefix = "room"
         session_id = f"{session_prefix}%{room_id}"
-        if unique_session:
-            session_id += f"_{sender_info['sender_id']}"
         message.type = MessageType.GROUP_MESSAGE
         message.group_id = room_id
     elif is_chat:

astrbot/core/platform/sources/qqofficial/qqofficial_platform_adapter.py

@@ -44,11 +44,8 @@ class botClient(Client):
             message,
             MessageType.GROUP_MESSAGE,
         )
-        abm.session_id = (
-            abm.sender.user_id
-            if self.platform.unique_session
-            else cast(str, message.group_openid)
-        )
+        abm.group_id = cast(str, message.group_openid)
+        abm.session_id = abm.group_id
         self._commit(abm)
 
     # 收到频道消息
@@ -57,9 +54,8 @@ class botClient(Client):
             message,
             MessageType.GROUP_MESSAGE,
         )
-        abm.session_id = (
-            abm.sender.user_id if self.platform.unique_session else message.channel_id
-        )
+        abm.group_id = message.channel_id
+        abm.session_id = abm.group_id
         self._commit(abm)
 
     # 收到私聊消息
@@ -104,7 +100,6 @@ class QQOfficialPlatformAdapter(Platform):
 
         self.appid = platform_config["appid"]
         self.secret = platform_config["secret"]
-        self.unique_session: bool = platform_settings["unique_session"]
         qq_group = platform_config["enable_group_c2c"]
         guild_dm = platform_config["enable_guild_direct_message"]
 

astrbot/core/platform/sources/qqofficial_webhook/qo_webhook_adapter.py

@@ -35,11 +35,8 @@ class botClient(Client):
             message,
             MessageType.GROUP_MESSAGE,
         )
-        abm.session_id = (
-            abm.sender.user_id
-            if self.platform.unique_session
-            else cast(str, message.group_openid)
-        )
+        abm.group_id = cast(str, message.group_openid)
+        abm.session_id = abm.group_id
         self._commit(abm)
 
     # 收到频道消息
@@ -48,9 +45,8 @@ class botClient(Client):
             message,
             MessageType.GROUP_MESSAGE,
         )
-        abm.session_id = (
-            abm.sender.user_id if self.platform.unique_session else message.channel_id
-        )
+        abm.group_id = message.channel_id
+        abm.session_id = abm.group_id
         self._commit(abm)
 
     # 收到私聊消息
@@ -95,7 +91,6 @@ class QQOfficialWebhookPlatformAdapter(Platform):
 
         self.appid = platform_config["appid"]
         self.secret = platform_config["secret"]
-        self.unique_session = platform_settings["unique_session"]
         self.unified_webhook_mode = platform_config.get("unified_webhook_mode", False)
 
         intents = botpy.Intents(

astrbot/core/platform/sources/satori/satori_adapter.py

@@ -142,7 +142,12 @@ class SatoriPlatformAdapter(Platform):
             raise ValueError(f"WebSocket URL必须以ws://或wss://开头: {self.endpoint}")
 
         try:
-            websocket = await connect(self.endpoint, additional_headers={})
+            websocket = await connect(
+                self.endpoint,
+                additional_headers={},
+                max_size=10 * 1024 * 1024,  # 10MB
+            )
+
             self.ws = websocket
 
             await asyncio.sleep(0.1)

astrbot/core/platform/sources/slack/slack_adapter.py

@@ -41,7 +41,6 @@ class SlackAdapter(Platform):
     ) -> None:
         super().__init__(platform_config, event_queue)
         self.settings = platform_settings
-        self.unique_session = platform_settings.get("unique_session", False)
 
         self.bot_token = platform_config.get("bot_token")
         self.app_token = platform_config.get("app_token")
@@ -147,12 +146,10 @@ class SlackAdapter(Platform):
             abm.group_id = channel_id
 
         # 设置会话ID
-        if self.unique_session and abm.type == MessageType.GROUP_MESSAGE:
-            abm.session_id = f"{user_id}_{channel_id}"
+        if abm.type == MessageType.GROUP_MESSAGE:
+            abm.session_id = abm.group_id
         else:
-            abm.session_id = (
-                channel_id if abm.type == MessageType.GROUP_MESSAGE else user_id
-            )
+            abm.session_id = user_id
 
         abm.message_id = event.get("client_msg_id", uuid.uuid4().hex)
         abm.timestamp = int(float(event.get("ts", time.time())))

astrbot/core/platform/sources/webchat/webchat_adapter.py

@@ -79,7 +79,6 @@ class WebChatAdapter(Platform):
         super().__init__(platform_config, event_queue)
 
         self.settings = platform_settings
-        self.unique_session = platform_settings["unique_session"]
         self.imgs_dir = os.path.join(get_astrbot_data_path(), "webchat", "imgs")
         os.makedirs(self.imgs_dir, exist_ok=True)
 

astrbot/core/platform/sources/wechatpadpro/wechatpadpro_adapter.py

@@ -47,7 +47,6 @@ class WeChatPadProAdapter(Platform):
         self._shutdown_event = None
         self.wxnewpass = None
         self.settings = platform_settings
-        self.unique_session = platform_settings.get("unique_session", False)
 
         self.metadata = PlatformMetadata(
             name="wechatpadpro",
@@ -509,11 +508,10 @@ class WeChatPadProAdapter(Platform):
                 if accurate_nickname:
                     abm.sender.nickname = accurate_nickname
 
-            # 对于群聊，session_id 可以是群聊 ID 或发送者 ID + 群聊 ID (如果 unique_session 为 True)
-            if self.unique_session:
-                abm.session_id = f"{from_user_name}#{abm.sender.user_id}"
+            if abm.type == MessageType.GROUP_MESSAGE:
+                abm.session_id = abm.group_id
             else:
-                abm.session_id = from_user_name
+                abm.session_id = abm.sender.user_id
 
             msg_source = raw_message.get("msg_source", "")
             if self.wxid in msg_source:

astrbot/core/platform/sources/weixin_official_account/weixin_offacc_adapter.py

@@ -191,7 +191,7 @@ class WeixinOfficialAccountPlatformAdapter(Platform):
                 if self.active_send_mode:
                     await self.convert_message(msg, None)
                 else:
-                    if msg.id in self.wexin_event_workers:
+                    if str(msg.id) in self.wexin_event_workers:
                         future = self.wexin_event_workers[str(cast(str | int, msg.id))]
                         logger.debug(f"duplicate message id checked: {msg.id}")
                     else:

astrbot/core/provider/entities.py

@@ -14,6 +14,7 @@ import astrbot.core.message.components as Comp
 from astrbot import logger
 from astrbot.core.agent.message import (
     AssistantMessageSegment,
+    ContentPart,
     ToolCall,
     ToolCallMessageSegment,
 )
@@ -92,6 +93,8 @@ class ProviderRequest:
     """会话 ID"""
     image_urls: list[str] = field(default_factory=list)
     """图片 URL 列表"""
+    extra_user_content_parts: list[ContentPart] = field(default_factory=list)
+    """额外的用户消息内容部分列表，用于在用户消息后添加额外的内容块（如系统提醒、指令等）。支持 dict 或 ContentPart 对象"""
     func_tool: ToolSet | None = None
     """可用的函数工具"""
     contexts: list[dict] = field(default_factory=list)
@@ -166,13 +169,23 @@ class ProviderRequest:
 
     async def assemble_context(self) -> dict:
         """将请求(prompt 和 image_urls)包装成 OpenAI 的消息格式。"""
+        # 构建内容块列表
+        content_blocks = []
+
+        # 1. 用户原始发言（OpenAI 建议：用户发言在前）
+        if self.prompt and self.prompt.strip():
+            content_blocks.append({"type": "text", "text": self.prompt})
+        elif self.image_urls:
+            # 如果没有文本但有图片，添加占位文本
+            content_blocks.append({"type": "text", "text": "[图片]"})
+
+        # 2. 额外的内容块（系统提醒、指令等）
+        if self.extra_user_content_parts:
+            for part in self.extra_user_content_parts:
+                content_blocks.append(part.model_dump())
+
+        # 3. 图片内容
         if self.image_urls:
-            user_content = {
-                "role": "user",
-                "content": [
-                    {"type": "text", "text": self.prompt if self.prompt else "[图片]"},
-                ],
-            }
             for image_url in self.image_urls:
                 if image_url.startswith("http"):
                     image_path = await download_image_by_url(image_url)
@@ -185,11 +198,21 @@ class ProviderRequest:
                 if not image_data:
                     logger.warning(f"图片 {image_url} 得到的结果为空，将忽略。")
                     continue
-                user_content["content"].append(
+                content_blocks.append(
                     {"type": "image_url", "image_url": {"url": image_data}},
                 )
-            return user_content
-        return {"role": "user", "content": self.prompt}
+
+        # 只有当只有一个来自 prompt 的文本块且没有额外内容块时，才降级为简单格式以保持向后兼容
+        if (
+            len(content_blocks) == 1
+            and content_blocks[0]["type"] == "text"
+            and not self.extra_user_content_parts
+            and not self.image_urls
+        ):
+            return {"role": "user", "content": content_blocks[0]["text"]}
+
+        # 否则返回多模态格式
+        return {"role": "user", "content": content_blocks}
 
     async def _encode_image_bs64(self, image_url: str) -> str:
         """将图片转换为 base64"""
@@ -249,6 +272,8 @@ class LLMResponse:
     """Tool call extra content. tool_call_id -> extra_content dict"""
     reasoning_content: str = ""
     """The reasoning content extracted from the LLM, if any."""
+    reasoning_signature: str | None = None
+    """The signature of the reasoning content, if any."""
 
     raw_completion: (
         ChatCompletion | GenerateContentResponse | AnthropicMessage | None
@@ -269,12 +294,14 @@ class LLMResponse:
     def __init__(
         self,
         role: str,
-        completion_text: str = "",
+        completion_text: str | None = None,
         result_chain: MessageChain | None = None,
         tools_call_args: list[dict[str, Any]] | None = None,
         tools_call_name: list[str] | None = None,
         tools_call_ids: list[str] | None = None,
         tools_call_extra_content: dict[str, dict[str, Any]] | None = None,
+        reasoning_content: str | None = None,
+        reasoning_signature: str | None = None,
         raw_completion: ChatCompletion
         | GenerateContentResponse
         | AnthropicMessage
@@ -294,6 +321,8 @@ class LLMResponse:
             raw_completion (ChatCompletion, optional): 原始响应, OpenAI 格式. Defaults to None.
 
         """
+        if reasoning_content is None:
+            reasoning_content = ""
         if tools_call_args is None:
             tools_call_args = []
         if tools_call_name is None:
@@ -310,6 +339,8 @@ class LLMResponse:
         self.tools_call_name = tools_call_name
         self.tools_call_ids = tools_call_ids
         self.tools_call_extra_content = tools_call_extra_content
+        self.reasoning_content = reasoning_content
+        self.reasoning_signature = reasoning_signature
         self.raw_completion = raw_completion
         self.is_chunk = is_chunk
 

astrbot/core/provider/manager.py

@@ -119,19 +119,34 @@ class ProviderManager:
             TTSProvider,
         ):
             self.curr_tts_provider_inst = prov
-            sp.put("curr_provider_tts", provider_id, scope="global", scope_id="global")
+            await sp.put_async(
+                key="curr_provider_tts",
+                value=provider_id,
+                scope="global",
+                scope_id="global",
+            )
         elif provider_type == ProviderType.SPEECH_TO_TEXT and isinstance(
             prov,
             STTProvider,
         ):
             self.curr_stt_provider_inst = prov
-            sp.put("curr_provider_stt", provider_id, scope="global", scope_id="global")
+            await sp.put_async(
+                key="curr_provider_stt",
+                value=provider_id,
+                scope="global",
+                scope_id="global",
+            )
         elif provider_type == ProviderType.CHAT_COMPLETION and isinstance(
             prov,
             Provider,
         ):
             self.curr_provider_inst = prov
-            sp.put("curr_provider", provider_id, scope="global", scope_id="global")
+            await sp.put_async(
+                key="curr_provider",
+                value=provider_id,
+                scope="global",
+                scope_id="global",
+            )
 
     async def get_provider_by_id(self, provider_id: str) -> Providers | None:
         """根据提供商 ID 获取提供商实例"""
@@ -206,21 +221,21 @@ class ProviderManager:
                 logger.error(traceback.format_exc())
                 logger.error(e)
 
-        selected_provider_id = sp.get(
-            "curr_provider",
-            self.provider_settings.get("default_provider_id"),
+        selected_provider_id = await sp.get_async(
+            key="curr_provider",
+            default=self.provider_settings.get("default_provider_id"),
             scope="global",
             scope_id="global",
         )
-        selected_stt_provider_id = sp.get(
-            "curr_provider_stt",
-            self.provider_stt_settings.get("provider_id"),
+        selected_stt_provider_id = await sp.get_async(
+            key="curr_provider_stt",
+            default=self.provider_stt_settings.get("provider_id"),
             scope="global",
             scope_id="global",
         )
-        selected_tts_provider_id = sp.get(
-            "curr_provider_tts",
-            self.provider_tts_settings.get("provider_id"),
+        selected_tts_provider_id = await sp.get_async(
+            key="curr_provider_tts",
+            default=self.provider_tts_settings.get("provider_id"),
             scope="global",
             scope_id="global",
         )

astrbot/core/provider/provider.py

@@ -4,7 +4,7 @@ import os
 from collections.abc import AsyncGenerator
 from typing import TypeAlias, Union
 
-from astrbot.core.agent.message import Message
+from astrbot.core.agent.message import ContentPart, Message
 from astrbot.core.agent.tool import ToolSet
 from astrbot.core.provider.entities import (
     LLMResponse,
@@ -103,6 +103,7 @@ class Provider(AbstractProvider):
         system_prompt: str | None = None,
         tool_calls_result: ToolCallsResult | list[ToolCallsResult] | None = None,
         model: str | None = None,
+        extra_user_content_parts: list[ContentPart] | None = None,
         **kwargs,
     ) -> LLMResponse:
         """获得 LLM 的文本对话结果。会使用当前的模型进行对话。
@@ -114,6 +115,7 @@ class Provider(AbstractProvider):
             tools: tool set
             contexts: 上下文，和 prompt 二选一使用
             tool_calls_result: 回传给 LLM 的工具调用结果。参考: https://platform.openai.com/docs/guides/function-calling
+            extra_user_content_parts: 额外的内容块列表，用于在用户消息后添加额外的文本块（如系统提醒、指令等）
             kwargs: 其他参数
 
         Notes:

astrbot/core/provider/sources/anthropic_source.py

@@ -11,6 +11,7 @@ from anthropic.types.usage import Usage
 
 from astrbot import logger
 from astrbot.api.provider import Provider
+from astrbot.core.agent.message import ContentPart, ImageURLPart, TextPart
 from astrbot.core.provider.entities import LLMResponse, TokenUsage
 from astrbot.core.provider.func_tool_manager import ToolSet
 from astrbot.core.utils.io import download_image_by_url
@@ -47,6 +48,8 @@ class ProviderAnthropic(Provider):
             base_url=self.base_url,
         )
 
+        self.thinking_config = provider_config.get("anth_thinking_config", {})
+
         self.set_model(provider_config.get("model", "unknown"))
 
     def _prepare_payload(self, messages: list[dict]):
@@ -63,12 +66,33 @@ class ProviderAnthropic(Provider):
         new_messages = []
         for message in messages:
             if message["role"] == "system":
-                system_prompt = message["content"]
+                system_prompt = message["content"] or "<empty system prompt>"
             elif message["role"] == "assistant":
                 blocks = []
-                if isinstance(message["content"], str):
+                reasoning_content = ""
+                thinking_signature = ""
+                if isinstance(message["content"], str) and message["content"].strip():
                     blocks.append({"type": "text", "text": message["content"]})
-                if "tool_calls" in message:
+                elif isinstance(message["content"], list):
+                    for part in message["content"]:
+                        if part.get("type") == "think":
+                            # only pick the last think part for now
+                            reasoning_content = part.get("think")
+                            thinking_signature = part.get("encrypted")
+                        else:
+                            blocks.append(part)
+
+                if reasoning_content and thinking_signature:
+                    blocks.insert(
+                        0,
+                        {
+                            "type": "thinking",
+                            "thinking": reasoning_content,
+                            "signature": thinking_signature,
+                        },
+                    )
+
+                if "tool_calls" in message and isinstance(message["tool_calls"], list):
                     for tool_call in message["tool_calls"]:
                         blocks.append(  # noqa: PERF401
                             {
@@ -99,7 +123,7 @@ class ProviderAnthropic(Provider):
                             {
                                 "type": "tool_result",
                                 "tool_use_id": message["tool_call_id"],
-                                "content": message["content"],
+                                "content": message["content"] or "<empty response>",
                             },
                         ],
                     },
@@ -132,6 +156,14 @@ class ProviderAnthropic(Provider):
 
         extra_body = self.provider_config.get("custom_extra_body", {})
 
+        if "max_tokens" not in payloads:
+            payloads["max_tokens"] = 1024
+        if self.thinking_config.get("budget"):
+            payloads["thinking"] = {
+                "budget_tokens": self.thinking_config.get("budget"),
+                "type": "enabled",
+            }
+
         completion = await self.client.messages.create(
             **payloads, stream=False, extra_body=extra_body
         )
@@ -149,6 +181,11 @@ class ProviderAnthropic(Provider):
                 completion_text = str(content_block.text).strip()
                 llm_response.completion_text = completion_text
 
+            if content_block.type == "thinking":
+                reasoning_content = str(content_block.thinking).strip()
+                llm_response.reasoning_content = reasoning_content
+                llm_response.reasoning_signature = content_block.signature
+
             if content_block.type == "tool_use":
                 llm_response.tools_call_args.append(content_block.input)
                 llm_response.tools_call_name.append(content_block.name)
@@ -180,6 +217,16 @@ class ProviderAnthropic(Provider):
         id = None
         usage = TokenUsage()
         extra_body = self.provider_config.get("custom_extra_body", {})
+        reasoning_content = ""
+        reasoning_signature = ""
+
+        if "max_tokens" not in payloads:
+            payloads["max_tokens"] = 1024
+        if self.thinking_config.get("budget"):
+            payloads["thinking"] = {
+                "budget_tokens": self.thinking_config.get("budget"),
+                "type": "enabled",
+            }
 
         async with self.client.messages.stream(
             **payloads, extra_body=extra_body
@@ -219,6 +266,21 @@ class ProviderAnthropic(Provider):
                             usage=usage,
                             id=id,
                         )
+                    elif event.delta.type == "thinking_delta":
+                        # 思考增量
+                        reasoning = event.delta.thinking
+                        if reasoning:
+                            yield LLMResponse(
+                                role="assistant",
+                                reasoning_content=reasoning,
+                                is_chunk=True,
+                                usage=usage,
+                                id=id,
+                                reasoning_signature=reasoning_signature or None,
+                            )
+                            reasoning_content += reasoning
+                    elif event.delta.type == "signature_delta":
+                        reasoning_signature = event.delta.signature
                     elif event.delta.type == "input_json_delta":
                         # 工具调用参数增量
                         if event.index in tool_use_buffer:
@@ -275,6 +337,8 @@ class ProviderAnthropic(Provider):
             is_chunk=False,
             usage=usage,
             id=id,
+            reasoning_content=reasoning_content,
+            reasoning_signature=reasoning_signature or None,
         )
 
         if final_tool_calls:
@@ -296,13 +360,16 @@ class ProviderAnthropic(Provider):
         system_prompt=None,
         tool_calls_result=None,
         model=None,
+        extra_user_content_parts=None,
         **kwargs,
     ) -> LLMResponse:
         if contexts is None:
             contexts = []
         new_record = None
         if prompt is not None:
-            new_record = await self.assemble_context(prompt, image_urls)
+            new_record = await self.assemble_context(
+                prompt, image_urls, extra_user_content_parts
+            )
         context_query = self._ensure_message_to_dicts(contexts)
         if new_record:
             context_query.append(new_record)
@@ -342,21 +409,24 @@ class ProviderAnthropic(Provider):
 
     async def text_chat_stream(
         self,
-        prompt,
+        prompt=None,
         session_id=None,
-        image_urls=...,
+        image_urls=None,
         func_tool=None,
-        contexts=...,
+        contexts=None,
         system_prompt=None,
         tool_calls_result=None,
         model=None,
+        extra_user_content_parts=None,
         **kwargs,
     ):
         if contexts is None:
             contexts = []
         new_record = None
         if prompt is not None:
-            new_record = await self.assemble_context(prompt, image_urls)
+            new_record = await self.assemble_context(
+                prompt, image_urls, extra_user_content_parts
+            )
         context_query = self._ensure_message_to_dicts(contexts)
         if new_record:
             context_query.append(new_record)
@@ -388,15 +458,15 @@ class ProviderAnthropic(Provider):
         async for llm_response in self._query_stream(payloads, func_tool):
             yield llm_response
 
-    async def assemble_context(self, text: str, image_urls: list[str] | None = None):
+    async def assemble_context(
+        self,
+        text: str,
+        image_urls: list[str] | None = None,
+        extra_user_content_parts: list[ContentPart] | None = None,
+    ):
         """组装上下文，支持文本和图片"""
-        if not image_urls:
-            return {"role": "user", "content": text}
 
-        content = []
-        content.append({"type": "text", "text": text})
-
-        for image_url in image_urls:
+        async def resolve_image_url(image_url: str) -> dict | None:
             if image_url.startswith("http"):
                 image_path = await download_image_by_url(image_url)
                 image_data = await self.encode_image_bs64(image_path)
@@ -408,28 +478,68 @@ class ProviderAnthropic(Provider):
 
             if not image_data:
                 logger.warning(f"图片 {image_url} 得到的结果为空，将忽略。")
-                continue
+                return None
 
             # Get mime type for the image
             mime_type, _ = guess_type(image_url)
             if not mime_type:
                 mime_type = "image/jpeg"  # Default to JPEG if can't determine
 
-            content.append(
-                {
-                    "type": "image",
-                    "source": {
-                        "type": "base64",
-                        "media_type": mime_type,
-                        "data": (
-                            image_data.split("base64,")[1]
-                            if "base64," in image_data
-                            else image_data
-                        ),
-                    },
+            return {
+                "type": "image",
+                "source": {
+                    "type": "base64",
+                    "media_type": mime_type,
+                    "data": (
+                        image_data.split("base64,")[1]
+                        if "base64," in image_data
+                        else image_data
+                    ),
                 },
-            )
+            }
 
+        content = []
+
+        # 1. 用户原始发言（OpenAI 建议：用户发言在前）
+        if text:
+            content.append({"type": "text", "text": text})
+        elif image_urls:
+            # 如果没有文本但有图片，添加占位文本
+            content.append({"type": "text", "text": "[图片]"})
+        elif extra_user_content_parts:
+            # 如果只有额外内容块，也需要添加占位文本
+            content.append({"type": "text", "text": " "})
+
+        # 2. 额外的内容块（系统提醒、指令等）
+        if extra_user_content_parts:
+            for block in extra_user_content_parts:
+                if isinstance(block, TextPart):
+                    content.append({"type": "text", "text": block.text})
+                elif isinstance(block, ImageURLPart):
+                    image_dict = await resolve_image_url(block.image_url.url)
+                    if image_dict:
+                        content.append(image_dict)
+                else:
+                    raise ValueError(f"不支持的额外内容块类型: {type(block)}")
+
+        # 3. 图片内容
+        if image_urls:
+            for image_url in image_urls:
+                image_dict = await resolve_image_url(image_url)
+                if image_dict:
+                    content.append(image_dict)
+
+        # 如果只有主文本且没有额外内容块和图片，返回简单格式以保持向后兼容
+        if (
+            text
+            and not extra_user_content_parts
+            and not image_urls
+            and len(content) == 1
+            and content[0]["type"] == "text"
+        ):
+            return {"role": "user", "content": content[0]["text"]}
+
+        # 否则返回多模态格式
         return {"role": "user", "content": content}
 
     async def encode_image_bs64(self, image_url: str) -> str:

astrbot/core/provider/sources/fishaudio_tts_api_source.py

@@ -56,10 +56,14 @@ class ProviderFishAudioTTSAPI(TTSProvider):
             "api_base",
             "https://api.fish-audio.cn/v1",
         )
+        try:
+            self.timeout: int = int(provider_config.get("timeout", 20))
+        except ValueError:
+            self.timeout = 20
         self.headers = {
             "Authorization": f"Bearer {self.chosen_api_key}",
         }
-        self.set_model(provider_config["model"])
+        self.set_model(provider_config.get("model", None))
 
     async def _get_reference_id_by_character(self, character: str) -> str | None:
         """获取角色的reference_id
@@ -135,17 +139,21 @@ class ProviderFishAudioTTSAPI(TTSProvider):
         path = os.path.join(temp_dir, f"fishaudio_tts_api_{uuid.uuid4()}.wav")
         self.headers["content-type"] = "application/msgpack"
         request = await self._generate_request(text)
-        async with AsyncClient(base_url=self.api_base).stream(
+        async with AsyncClient(base_url=self.api_base, timeout=self.timeout).stream(
             "POST",
             "/tts",
             headers=self.headers,
             content=ormsgpack.packb(request, option=ormsgpack.OPT_SERIALIZE_PYDANTIC),
         ) as response:
-            if response.headers["content-type"] == "audio/wav":
+            if response.status_code == 200 and response.headers.get(
+                "content-type", ""
+            ).startswith("audio/"):
                 with open(path, "wb") as f:
                     async for chunk in response.aiter_bytes():
                         f.write(chunk)
                 return path
-            body = await response.aread()
-            text = body.decode("utf-8", errors="replace")
-            raise Exception(f"Fish Audio API请求失败: {text}")
+            error_bytes = await response.aread()
+            error_text = error_bytes.decode("utf-8", errors="replace")[:1024]
+            raise Exception(
+                f"Fish Audio API请求失败: 状态码 {response.status_code}, 响应内容: {error_text}"
+            )

astrbot/core/provider/sources/gemini_source.py

@@ -13,6 +13,7 @@ from google.genai.errors import APIError
 import astrbot.core.message.components as Comp
 from astrbot import logger
 from astrbot.api.provider import Provider
+from astrbot.core.agent.message import ContentPart, ImageURLPart, TextPart
 from astrbot.core.message.message_event_result import MessageChain
 from astrbot.core.provider.entities import LLMResponse, TokenUsage
 from astrbot.core.provider.func_tool_manager import ToolSet
@@ -320,9 +321,37 @@ class ProviderGoogleGenAI(Provider):
                 append_or_extend(gemini_contents, parts, types.UserContent)
 
             elif role == "assistant":
-                if content:
+                if isinstance(content, str):
                     parts = [types.Part.from_text(text=content)]
                     append_or_extend(gemini_contents, parts, types.ModelContent)
+                elif isinstance(content, list):
+                    parts = []
+                    thinking_signature = None
+                    text = ""
+                    for part in content:
+                        # for most cases, assistant content only contains two parts: think and text
+                        if part.get("type") == "think":
+                            thinking_signature = part.get("encrypted") or None
+                        else:
+                            text += str(part.get("text"))
+
+                    if thinking_signature and isinstance(thinking_signature, str):
+                        try:
+                            thinking_signature = base64.b64decode(thinking_signature)
+                        except Exception as e:
+                            logger.warning(
+                                f"Failed to decode google gemini thinking signature: {e}",
+                                exc_info=True,
+                            )
+                            thinking_signature = None
+                    parts.append(
+                        types.Part(
+                            text=text,
+                            thought_signature=thinking_signature,
+                        )
+                    )
+                    append_or_extend(gemini_contents, parts, types.ModelContent)
+
                 elif not native_tool_enabled and "tool_calls" in message:
                     parts = []
                     for tool in message["tool_calls"]:
@@ -440,7 +469,8 @@ class ProviderGoogleGenAI(Provider):
         for part in result_parts:
             if part.text:
                 chain.append(Comp.Plain(part.text))
-            elif (
+
+            if (
                 part.function_call
                 and part.function_call.name is not None
                 and part.function_call.args is not None
@@ -457,13 +487,18 @@ class ProviderGoogleGenAI(Provider):
                     llm_response.tools_call_extra_content[tool_call_id] = {
                         "google": {"thought_signature": ts_bs64}
                     }
-            elif (
+
+            if (
                 part.inline_data
                 and part.inline_data.mime_type
                 and part.inline_data.mime_type.startswith("image/")
                 and part.inline_data.data
             ):
                 chain.append(Comp.Image.fromBytes(part.inline_data.data))
+
+            if ts := part.thought_signature:
+                # only keep the last thinking signature
+                llm_response.reasoning_signature = base64.b64encode(ts).decode("utf-8")
         return MessageChain(chain=chain)
 
     async def _query(self, payloads: dict, tools: ToolSet | None) -> LLMResponse:
@@ -680,13 +715,16 @@ class ProviderGoogleGenAI(Provider):
         system_prompt=None,
         tool_calls_result=None,
         model=None,
+        extra_user_content_parts=None,
         **kwargs,
     ) -> LLMResponse:
         if contexts is None:
             contexts = []
         new_record = None
         if prompt is not None:
-            new_record = await self.assemble_context(prompt, image_urls)
+            new_record = await self.assemble_context(
+                prompt, image_urls, extra_user_content_parts
+            )
         context_query = self._ensure_message_to_dicts(contexts)
         if new_record:
             context_query.append(new_record)
@@ -732,13 +770,16 @@ class ProviderGoogleGenAI(Provider):
         system_prompt=None,
         tool_calls_result=None,
         model=None,
+        extra_user_content_parts=None,
         **kwargs,
     ) -> AsyncGenerator[LLMResponse, None]:
         if contexts is None:
             contexts = []
         new_record = None
         if prompt is not None:
-            new_record = await self.assemble_context(prompt, image_urls)
+            new_record = await self.assemble_context(
+                prompt, image_urls, extra_user_content_parts
+            )
         context_query = self._ensure_message_to_dicts(contexts)
         if new_record:
             context_query.append(new_record)
@@ -797,33 +838,75 @@ class ProviderGoogleGenAI(Provider):
         self.chosen_api_key = key
         self._init_client()
 
-    async def assemble_context(self, text: str, image_urls: list[str] | None = None):
+    async def assemble_context(
+        self,
+        text: str,
+        image_urls: list[str] | None = None,
+        extra_user_content_parts: list[ContentPart] | None = None,
+    ):
         """组装上下文。"""
-        if image_urls:
-            user_content = {
-                "role": "user",
-                "content": [{"type": "text", "text": text if text else "[图片]"}],
+
+        async def resolve_image_part(image_url: str) -> dict | None:
+            if image_url.startswith("http"):
+                image_path = await download_image_by_url(image_url)
+                image_data = await self.encode_image_bs64(image_path)
+            elif image_url.startswith("file:///"):
+                image_path = image_url.replace("file:///", "")
+                image_data = await self.encode_image_bs64(image_path)
+            else:
+                image_data = await self.encode_image_bs64(image_url)
+            if not image_data:
+                logger.warning(f"图片 {image_url} 得到的结果为空，将忽略。")
+                return None
+            return {
+                "type": "image_url",
+                "image_url": {"url": image_data},
             }
-            for image_url in image_urls:
-                if image_url.startswith("http"):
-                    image_path = await download_image_by_url(image_url)
-                    image_data = await self.encode_image_bs64(image_path)
-                elif image_url.startswith("file:///"):
-                    image_path = image_url.replace("file:///", "")
-                    image_data = await self.encode_image_bs64(image_path)
+
+        # 构建内容块列表
+        content_blocks = []
+
+        # 1. 用户原始发言（OpenAI 建议：用户发言在前）
+        if text:
+            content_blocks.append({"type": "text", "text": text})
+        elif image_urls:
+            # 如果没有文本但有图片，添加占位文本
+            content_blocks.append({"type": "text", "text": "[图片]"})
+        elif extra_user_content_parts:
+            # 如果只有额外内容块，也需要添加占位文本
+            content_blocks.append({"type": "text", "text": " "})
+
+        # 2. 额外的内容块（系统提醒、指令等）
+        if extra_user_content_parts:
+            for part in extra_user_content_parts:
+                if isinstance(part, TextPart):
+                    content_blocks.append({"type": "text", "text": part.text})
+                elif isinstance(part, ImageURLPart):
+                    image_part = await resolve_image_part(part.image_url.url)
+                    if image_part:
+                        content_blocks.append(image_part)
                 else:
-                    image_data = await self.encode_image_bs64(image_url)
-                if not image_data:
-                    logger.warning(f"图片 {image_url} 得到的结果为空，将忽略。")
-                    continue
-                user_content["content"].append(
-                    {
-                        "type": "image_url",
-                        "image_url": {"url": image_data},
-                    },
-                )
-            return user_content
-        return {"role": "user", "content": text}
+                    raise ValueError(f"不支持的额外内容块类型: {type(part)}")
+
+        # 3. 图片内容
+        if image_urls:
+            for image_url in image_urls:
+                image_part = await resolve_image_part(image_url)
+                if image_part:
+                    content_blocks.append(image_part)
+
+        # 如果只有主文本且没有额外内容块和图片，返回简单格式以保持向后兼容
+        if (
+            text
+            and not extra_user_content_parts
+            and not image_urls
+            and len(content_blocks) == 1
+            and content_blocks[0]["type"] == "text"
+        ):
+            return {"role": "user", "content": content_blocks[0]["text"]}
+
+        # 否则返回多模态格式
+        return {"role": "user", "content": content_blocks}
 
     async def encode_image_bs64(self, image_url: str) -> str:
         """将图片转换为 base64"""

astrbot/core/provider/sources/minimax_tts_api_source.py

@@ -51,7 +51,7 @@ class ProviderMiniMaxTTSAPI(TTSProvider):
             "voice_id": ""
             if self.is_timber_weight
             else provider_config.get("minimax-voice-id", ""),
-            "emotion": provider_config.get("minimax-voice-emotion", "neutral"),
+            "emotion": provider_config.get("minimax-voice-emotion", "auto"),
             "latex_read": provider_config.get("minimax-voice-latex", False),
             "english_normalization": provider_config.get(
                 "minimax-voice-english-normalization",
@@ -59,6 +59,9 @@ class ProviderMiniMaxTTSAPI(TTSProvider):
             ),
         }
 
+        if self.voice_setting["emotion"] == "auto":
+            self.voice_setting.pop("emotion", None)
+
         self.audio_setting: dict = {
             "sample_rate": 32000,
             "bitrate": 128000,

astrbot/core/provider/sources/openai_source.py

@@ -17,7 +17,7 @@ from openai.types.completion_usage import CompletionUsage
 import astrbot.core.message.components as Comp
 from astrbot import logger
 from astrbot.api.provider import Provider
-from astrbot.core.agent.message import Message
+from astrbot.core.agent.message import ContentPart, ImageURLPart, Message, TextPart
 from astrbot.core.agent.tool import ToolSet
 from astrbot.core.message.message_event_result import MessageChain
 from astrbot.core.provider.entities import LLMResponse, TokenUsage, ToolCallsResult
@@ -74,28 +74,6 @@ class ProviderOpenAIOfficial(Provider):
 
         self.reasoning_key = "reasoning_content"
 
-    def _maybe_inject_xai_search(self, payloads: dict, **kwargs):
-        """当开启 xAI 原生搜索时，向请求体注入 Live Search 参数。
-
-        - 仅在 provider_config.xai_native_search 为 True 时生效
-        - 默认注入 {"mode": "auto"}
-        - 允许通过 kwargs 使用 xai_search_mode 覆盖（on/auto/off）
-        """
-        if not bool(self.provider_config.get("xai_native_search", False)):
-            return
-
-        mode = kwargs.get("xai_search_mode", "auto")
-        mode = str(mode).lower()
-        if mode not in ("auto", "on", "off"):
-            mode = "auto"
-
-        # off 时不注入，保持与未开启一致
-        if mode == "off":
-            return
-
-        # OpenAI SDK 不识别的字段会在 _query/_query_stream 中放入 extra_body
-        payloads["search_parameters"] = {"mode": mode}
-
     async def get_models(self):
         try:
             models_str = []
@@ -134,10 +112,6 @@ class ProviderOpenAIOfficial(Provider):
 
         model = payloads.get("model", "").lower()
 
-        # 针对 deepseek 模型的特殊处理：deepseek-reasoner调用必须移除 tools ，否则将被切换至 deepseek-chat
-        if model == "deepseek-reasoner" and "tools" in payloads:
-            del payloads["tools"]
-
         completion = await self.client.chat.completions.create(
             **payloads,
             stream=False,
@@ -251,10 +225,14 @@ class ProviderOpenAIOfficial(Provider):
     def _extract_usage(self, usage: CompletionUsage) -> TokenUsage:
         ptd = usage.prompt_tokens_details
         cached = ptd.cached_tokens if ptd and ptd.cached_tokens else 0
+        prompt_tokens = 0 if usage.prompt_tokens is None else usage.prompt_tokens
+        completion_tokens = (
+            0 if usage.completion_tokens is None else usage.completion_tokens
+        )
         return TokenUsage(
-            input_other=usage.prompt_tokens - cached,
-            input_cached=ptd.cached_tokens if ptd and ptd.cached_tokens else 0,
-            output=usage.completion_tokens,
+            input_other=prompt_tokens - cached,
+            input_cached=cached,
+            output=completion_tokens,
         )
 
     async def _parse_openai_completion(
@@ -348,6 +326,7 @@ class ProviderOpenAIOfficial(Provider):
         system_prompt: str | None = None,
         tool_calls_result: ToolCallsResult | list[ToolCallsResult] | None = None,
         model: str | None = None,
+        extra_user_content_parts: list[ContentPart] | None = None,
         **kwargs,
     ) -> tuple:
         """准备聊天所需的有效载荷和上下文"""
@@ -355,7 +334,9 @@ class ProviderOpenAIOfficial(Provider):
             contexts = []
         new_record = None
         if prompt is not None:
-            new_record = await self.assemble_context(prompt, image_urls)
+            new_record = await self.assemble_context(
+                prompt, image_urls, extra_user_content_parts
+            )
         context_query = self._ensure_message_to_dicts(contexts)
         if new_record:
             context_query.append(new_record)
@@ -378,11 +359,28 @@ class ProviderOpenAIOfficial(Provider):
 
         payloads = {"messages": context_query, "model": model}
 
-        # xAI origin search tool inject
-        self._maybe_inject_xai_search(payloads, **kwargs)
+        self._finally_convert_payload(payloads)
 
         return payloads, context_query
 
+    def _finally_convert_payload(self, payloads: dict):
+        """Finally convert the payload. Such as think part conversion, tool inject."""
+        for message in payloads.get("messages", []):
+            if message.get("role") == "assistant" and isinstance(
+                message.get("content"), list
+            ):
+                reasoning_content = ""
+                new_content = []  # not including think part
+                for part in message["content"]:
+                    if part.get("type") == "think":
+                        reasoning_content += str(part.get("think"))
+                    else:
+                        new_content.append(part)
+                message["content"] = new_content
+                # reasoning key is "reasoning_content"
+                if reasoning_content:
+                    message["reasoning_content"] = reasoning_content
+
     async def _handle_api_error(
         self,
         e: Exception,
@@ -476,6 +474,7 @@ class ProviderOpenAIOfficial(Provider):
         system_prompt=None,
         tool_calls_result=None,
         model=None,
+        extra_user_content_parts=None,
         **kwargs,
     ) -> LLMResponse:
         payloads, context_query = await self._prepare_chat_payload(
@@ -485,6 +484,7 @@ class ProviderOpenAIOfficial(Provider):
             system_prompt,
             tool_calls_result,
             model=model,
+            extra_user_content_parts=extra_user_content_parts,
             **kwargs,
         )
 
@@ -624,33 +624,71 @@ class ProviderOpenAIOfficial(Provider):
         self,
         text: str,
         image_urls: list[str] | None = None,
+        extra_user_content_parts: list[ContentPart] | None = None,
     ) -> dict:
         """组装成符合 OpenAI 格式的 role 为 user 的消息段"""
-        if image_urls:
-            user_content = {
-                "role": "user",
-                "content": [{"type": "text", "text": text if text else "[图片]"}],
+
+        async def resolve_image_part(image_url: str) -> dict | None:
+            if image_url.startswith("http"):
+                image_path = await download_image_by_url(image_url)
+                image_data = await self.encode_image_bs64(image_path)
+            elif image_url.startswith("file:///"):
+                image_path = image_url.replace("file:///", "")
+                image_data = await self.encode_image_bs64(image_path)
+            else:
+                image_data = await self.encode_image_bs64(image_url)
+            if not image_data:
+                logger.warning(f"图片 {image_url} 得到的结果为空，将忽略。")
+                return None
+            return {
+                "type": "image_url",
+                "image_url": {"url": image_data},
             }
-            for image_url in image_urls:
-                if image_url.startswith("http"):
-                    image_path = await download_image_by_url(image_url)
-                    image_data = await self.encode_image_bs64(image_path)
-                elif image_url.startswith("file:///"):
-                    image_path = image_url.replace("file:///", "")
-                    image_data = await self.encode_image_bs64(image_path)
+
+        # 构建内容块列表
+        content_blocks = []
+
+        # 1. 用户原始发言（OpenAI 建议：用户发言在前）
+        if text:
+            content_blocks.append({"type": "text", "text": text})
+        elif image_urls:
+            # 如果没有文本但有图片，添加占位文本
+            content_blocks.append({"type": "text", "text": "[图片]"})
+        elif extra_user_content_parts:
+            # 如果只有额外内容块，也需要添加占位文本
+            content_blocks.append({"type": "text", "text": " "})
+
+        # 2. 额外的内容块（系统提醒、指令等）
+        if extra_user_content_parts:
+            for part in extra_user_content_parts:
+                if isinstance(part, TextPart):
+                    content_blocks.append({"type": "text", "text": part.text})
+                elif isinstance(part, ImageURLPart):
+                    image_part = await resolve_image_part(part.image_url.url)
+                    if image_part:
+                        content_blocks.append(image_part)
                 else:
-                    image_data = await self.encode_image_bs64(image_url)
-                if not image_data:
-                    logger.warning(f"图片 {image_url} 得到的结果为空，将忽略。")
-                    continue
-                user_content["content"].append(
-                    {
-                        "type": "image_url",
-                        "image_url": {"url": image_data},
-                    },
-                )
-            return user_content
-        return {"role": "user", "content": text}
+                    raise ValueError(f"不支持的额外内容块类型: {type(part)}")
+
+        # 3. 图片内容
+        if image_urls:
+            for image_url in image_urls:
+                image_part = await resolve_image_part(image_url)
+                if image_part:
+                    content_blocks.append(image_part)
+
+        # 如果只有主文本且没有额外内容块和图片，返回简单格式以保持向后兼容
+        if (
+            text
+            and not extra_user_content_parts
+            and not image_urls
+            and len(content_blocks) == 1
+            and content_blocks[0]["type"] == "text"
+        ):
+            return {"role": "user", "content": content_blocks[0]["text"]}
+
+        # 否则返回多模态格式
+        return {"role": "user", "content": content_blocks}
 
     async def encode_image_bs64(self, image_url: str) -> str:
         """将图片转换为 base64"""

astrbot/core/provider/sources/xai_source.py

@@ -0,0 +1,29 @@
+from ..register import register_provider_adapter
+from .openai_source import ProviderOpenAIOfficial
+
+
+@register_provider_adapter(
+    "xai_chat_completion", "xAI Chat Completion Provider Adapter"
+)
+class ProviderXAI(ProviderOpenAIOfficial):
+    def __init__(
+        self,
+        provider_config: dict,
+        provider_settings: dict,
+    ) -> None:
+        super().__init__(provider_config, provider_settings)
+
+    def _maybe_inject_xai_search(self, payloads: dict):
+        """当开启 xAI 原生搜索时，向请求体注入 Live Search 参数。
+
+        - 仅在 provider_config.xai_native_search 为 True 时生效
+        - 默认注入 {"mode": "auto"}
+        """
+        if not bool(self.provider_config.get("xai_native_search", False)):
+            return
+        # OpenAI SDK 不识别的字段会在 _query/_query_stream 中放入 extra_body
+        payloads["search_parameters"] = {"mode": "auto"}
+
+    def _finally_convert_payload(self, payloads: dict):
+        self._maybe_inject_xai_search(payloads)
+        super()._finally_convert_payload(payloads)

astrbot/core/provider/sources/xinference_stt_provider.py

@@ -8,7 +8,10 @@ from xinference_client.client.restful.async_restful_client import (
 
 from astrbot.core import logger
 from astrbot.core.utils.astrbot_path import get_astrbot_data_path
-from astrbot.core.utils.tencent_record_helper import tencent_silk_to_wav
+from astrbot.core.utils.tencent_record_helper import (
+    convert_to_pcm_wav,
+    tencent_silk_to_wav,
+)
 
 from ..entities import ProviderType
 from ..provider import STTProvider
@@ -111,17 +114,22 @@ class ProviderXinferenceSTT(STTProvider):
                 return ""
 
             # 2. Check for conversion
-            needs_conversion = False
-            if (
-                audio_url.endswith((".amr", ".silk"))
-                or is_tencent
-                or b"SILK" in audio_bytes[:8]
-            ):
-                needs_conversion = True
+            conversion_type = None
+
+            if b"SILK" in audio_bytes[:8]:
+                conversion_type = "silk"
+            elif b"#!AMR" in audio_bytes[:6]:
+                conversion_type = "amr"
+            elif audio_url.endswith(".silk") or is_tencent:
+                conversion_type = "silk"
+            elif audio_url.endswith(".amr"):
+                conversion_type = "amr"
 
             # 3. Perform conversion if needed
-            if needs_conversion:
-                logger.info("Audio requires conversion, using temporary files...")
+            if conversion_type:
+                logger.info(
+                    f"Audio requires conversion ({conversion_type}), using temporary files..."
+                )
                 temp_dir = os.path.join(get_astrbot_data_path(), "temp")
                 os.makedirs(temp_dir, exist_ok=True)
 
@@ -132,8 +140,12 @@ class ProviderXinferenceSTT(STTProvider):
                 with open(input_path, "wb") as f:
                     f.write(audio_bytes)
 
-                logger.info("Converting silk/amr file to wav ...")
-                await tencent_silk_to_wav(input_path, output_path)
+                if conversion_type == "silk":
+                    logger.info("Converting silk to wav ...")
+                    await tencent_silk_to_wav(input_path, output_path)
+                elif conversion_type == "amr":
+                    logger.info("Converting amr to wav ...")
+                    await convert_to_pcm_wav(input_path, output_path)
 
                 with open(output_path, "rb") as f:
                     audio_bytes = f.read()

astrbot/core/star/command_management.py

@@ -4,7 +4,7 @@ from collections import defaultdict
 from dataclasses import dataclass, field
 from typing import Any
 
-from astrbot.core import db_helper
+from astrbot.core import db_helper, logger
 from astrbot.core.db.po import CommandConfig
 from astrbot.core.star.filter.command import CommandFilter
 from astrbot.core.star.filter.command_group import CommandGroupFilter
@@ -90,6 +90,7 @@ async def toggle_command(handler_full_name: str, enabled: bool) -> CommandDescri
 async def rename_command(
     handler_full_name: str,
     new_fragment: str,
+    aliases: list[str] | None = None,
 ) -> CommandDescriptor:
     descriptor = _build_descriptor_by_full_name(handler_full_name)
     if not descriptor:
@@ -99,9 +100,24 @@ async def rename_command(
     if not new_fragment:
         raise ValueError("指令名不能为空。")
 
+    # 校验主指令名
     candidate_full = _compose_command(descriptor.parent_signature, new_fragment)
     if _is_command_in_use(handler_full_name, candidate_full):
-        raise ValueError("新的指令名已被其他指令占用，请换一个名称。")
+        raise ValueError(f"指令名 '{candidate_full}' 已被其他指令占用。")
+
+    # 校验别名
+    if aliases:
+        for alias in aliases:
+            alias = alias.strip()
+            if not alias:
+                continue
+            alias_full = _compose_command(descriptor.parent_signature, alias)
+            if _is_command_in_use(handler_full_name, alias_full):
+                raise ValueError(f"别名 '{alias_full}' 已被其他指令占用。")
+
+    existing_cfg = await db_helper.get_command_config(handler_full_name)
+    merged_extra = dict(existing_cfg.extra_data or {}) if existing_cfg else {}
+    merged_extra["resolved_aliases"] = aliases or []
 
     config = await db_helper.upsert_command_config(
         handler_full_name=handler_full_name,
@@ -114,7 +130,7 @@ async def rename_command(
         conflict_key=descriptor.original_command,
         resolution_strategy="manual_rename",
         note=None,
-        extra_data=None,
+        extra_data=merged_extra,
         auto_managed=False,
     )
     _bind_descriptor_with_config(descriptor, config)
@@ -192,12 +208,18 @@ def _collect_descriptors(include_sub_commands: bool) -> list[CommandDescriptor]:
     """收集指令，按需包含子指令。"""
     descriptors: list[CommandDescriptor] = []
     for handler in star_handlers_registry:
-        desc = _build_descriptor(handler)
-        if not desc:
+        try:
+            desc = _build_descriptor(handler)
+            if not desc:
+                continue
+            if not include_sub_commands and desc.is_sub_command:
+                continue
+            descriptors.append(desc)
+        except Exception as e:
+            logger.warning(
+                f"解析指令处理函数 {handler.handler_full_name} 失败，跳过该指令。原因: {e!s}"
+            )
             continue
-        if not include_sub_commands and desc.is_sub_command:
-            continue
-        descriptors.append(desc)
     return descriptors
 
 
@@ -357,14 +379,27 @@ def _apply_config_to_descriptor(
         new_fragment,
     )
 
+    extra = config.extra_data or {}
+    resolved_aliases = extra.get("resolved_aliases")
+    if isinstance(resolved_aliases, list):
+        descriptor.aliases = [str(x) for x in resolved_aliases if str(x).strip()]
+
 
 def _apply_config_to_runtime(
     descriptor: CommandDescriptor,
     config: CommandConfig,
 ) -> None:
     descriptor.handler.enabled = config.enabled
-    if descriptor.filter_ref and descriptor.current_fragment:
-        _set_filter_fragment(descriptor.filter_ref, descriptor.current_fragment)
+    if descriptor.filter_ref:
+        if descriptor.current_fragment:
+            _set_filter_fragment(descriptor.filter_ref, descriptor.current_fragment)
+        extra = config.extra_data or {}
+        resolved_aliases = extra.get("resolved_aliases")
+        if isinstance(resolved_aliases, list):
+            _set_filter_aliases(
+                descriptor.filter_ref,
+                [str(x) for x in resolved_aliases if str(x).strip()],
+            )
 
 
 def _bind_configs_to_descriptors(
@@ -403,6 +438,18 @@ def _set_filter_fragment(
         filter_ref._cmpl_cmd_names = None
 
 
+def _set_filter_aliases(
+    filter_ref: CommandFilter | CommandGroupFilter,
+    aliases: list[str],
+) -> None:
+    current_aliases = getattr(filter_ref, "alias", set())
+    if set(aliases) == current_aliases:
+        return
+    setattr(filter_ref, "alias", set(aliases))
+    if hasattr(filter_ref, "_cmpl_cmd_names"):
+        filter_ref._cmpl_cmd_names = None
+
+
 def _is_command_in_use(
     target_handler_full_name: str,
     candidate_full_command: str,

astrbot/core/star/context.py

@@ -149,9 +149,12 @@ class Context:
             contexts: context messages for the LLM
             max_steps: Maximum number of tool calls before stopping the loop
             **kwargs: Additional keyword arguments. The kwargs will not be passed to the LLM directly for now, but can include:
+                stream: bool - whether to stream the LLM response
                 agent_hooks: BaseAgentRunHooks[AstrAgentContext] - hooks to run during agent execution
                 agent_context: AstrAgentContext - context to use for the agent
 
+                other kwargs will be DIRECTLY passed to the runner.reset() method
+
         Returns:
             The final LLMResponse after tool calls are completed.
 
@@ -194,6 +197,15 @@ class Context:
             )
         agent_runner = ToolLoopAgentRunner()
         tool_executor = FunctionToolExecutor()
+
+        streaming = kwargs.get("stream", False)
+
+        other_kwargs = {
+            k: v
+            for k, v in kwargs.items()
+            if k not in ["stream", "agent_hooks", "agent_context"]
+        }
+
         await agent_runner.reset(
             provider=prov,
             request=request,
@@ -203,7 +215,8 @@ class Context:
             ),
             tool_executor=tool_executor,
             agent_hooks=agent_hooks,
-            streaming=kwargs.get("stream", False),
+            streaming=streaming,
+            **other_kwargs,
         )
         async for _ in agent_runner.step_until_done(max_steps):
             pass
@@ -377,7 +390,7 @@ class Context:
             if not module_path:
                 _parts = []
                 module_part = tool.__module__.split(".")
-                flags = ["packages", "plugins"]
+                flags = ["builtin_stars", "plugins"]
                 for i, part in enumerate(module_part):
                     _parts.append(part)
                     if part in flags and i + 1 < len(module_part):

astrbot/core/star/register/__init__.py

@@ -12,6 +12,7 @@ from .star_handler import (
     register_on_llm_request,
     register_on_llm_response,
     register_on_platform_loaded,
+    register_on_waiting_llm_request,
     register_permission_type,
     register_platform_adapter_type,
     register_regex,
@@ -30,6 +31,7 @@ __all__ = [
     "register_on_llm_request",
     "register_on_llm_response",
     "register_on_platform_loaded",
+    "register_on_waiting_llm_request",
     "register_permission_type",
     "register_platform_adapter_type",
     "register_regex",

astrbot/core/star/register/star_handler.py

@@ -339,6 +339,30 @@ def register_on_platform_loaded(**kwargs):
     return decorator
 
 
+def register_on_waiting_llm_request(**kwargs):
+    """当等待调用 LLM 时的通知事件（在获取锁之前）
+
+    此钩子在消息确定要调用 LLM 但还未开始排队等锁时触发，
+    适合用于发送"正在思考中..."等用户反馈提示。
+
+    Examples:
+    ```py
+    @on_waiting_llm_request()
+    async def on_waiting_llm(self, event: AstrMessageEvent) -> None:
+        await event.send("🤔 正在思考中...")
+    ```
+
+    """
+
+    def decorator(awaitable):
+        _ = get_handler_or_create(
+            awaitable, EventType.OnWaitingLLMRequestEvent, **kwargs
+        )
+        return awaitable
+
+    return decorator
+
+
 def register_on_llm_request(**kwargs):
     """当有 LLM 请求时的事件
 

astrbot/core/star/session_llm_manager.py

@@ -12,7 +12,7 @@ class SessionServiceManager:
     # =============================================================================
 
     @staticmethod
-    def is_llm_enabled_for_session(session_id: str) -> bool:
+    async def is_llm_enabled_for_session(session_id: str) -> bool:
         """检查LLM是否在指定会话中启用
 
         Args:
@@ -23,11 +23,11 @@ class SessionServiceManager:
 
         """
         # 获取会话服务配置
-        session_services = sp.get(
-            "session_service_config",
-            {},
+        session_services = await sp.get_async(
             scope="umo",
             scope_id=session_id,
+            key="session_service_config",
+            default={},
         )
 
         # 如果配置了该会话的LLM状态，返回该状态
@@ -39,7 +39,7 @@ class SessionServiceManager:
         return True
 
     @staticmethod
-    def set_llm_status_for_session(session_id: str, enabled: bool) -> None:
+    async def set_llm_status_for_session(session_id: str, enabled: bool) -> None:
         """设置LLM在指定会话中的启停状态
 
         Args:
@@ -48,18 +48,24 @@ class SessionServiceManager:
 
         """
         session_config = (
-            sp.get("session_service_config", {}, scope="umo", scope_id=session_id) or {}
+            await sp.get_async(
+                scope="umo",
+                scope_id=session_id,
+                key="session_service_config",
+                default={},
+            )
+            or {}
         )
         session_config["llm_enabled"] = enabled
-        sp.put(
-            "session_service_config",
-            session_config,
+        await sp.put_async(
             scope="umo",
             scope_id=session_id,
+            key="session_service_config",
+            value=session_config,
         )
 
     @staticmethod
-    def should_process_llm_request(event: AstrMessageEvent) -> bool:
+    async def should_process_llm_request(event: AstrMessageEvent) -> bool:
         """检查是否应该处理LLM请求
 
         Args:
@@ -70,14 +76,14 @@ class SessionServiceManager:
 
         """
         session_id = event.unified_msg_origin
-        return SessionServiceManager.is_llm_enabled_for_session(session_id)
+        return await SessionServiceManager.is_llm_enabled_for_session(session_id)
 
     # =============================================================================
     # TTS 相关方法
     # =============================================================================
 
     @staticmethod
-    def is_tts_enabled_for_session(session_id: str) -> bool:
+    async def is_tts_enabled_for_session(session_id: str) -> bool:
         """检查TTS是否在指定会话中启用
 
         Args:
@@ -88,11 +94,11 @@ class SessionServiceManager:
 
         """
         # 获取会话服务配置
-        session_services = sp.get(
-            "session_service_config",
-            {},
+        session_services = await sp.get_async(
             scope="umo",
             scope_id=session_id,
+            key="session_service_config",
+            default={},
         )
 
         # 如果配置了该会话的TTS状态，返回该状态
@@ -104,7 +110,7 @@ class SessionServiceManager:
         return True
 
     @staticmethod
-    def set_tts_status_for_session(session_id: str, enabled: bool) -> None:
+    async def set_tts_status_for_session(session_id: str, enabled: bool) -> None:
         """设置TTS在指定会话中的启停状态
 
         Args:
@@ -113,14 +119,20 @@ class SessionServiceManager:
 
         """
         session_config = (
-            sp.get("session_service_config", {}, scope="umo", scope_id=session_id) or {}
+            await sp.get_async(
+                scope="umo",
+                scope_id=session_id,
+                key="session_service_config",
+                default={},
+            )
+            or {}
         )
         session_config["tts_enabled"] = enabled
-        sp.put(
-            "session_service_config",
-            session_config,
+        await sp.put_async(
             scope="umo",
             scope_id=session_id,
+            key="session_service_config",
+            value=session_config,
         )
 
         logger.info(
@@ -128,7 +140,7 @@ class SessionServiceManager:
         )
 
     @staticmethod
-    def should_process_tts_request(event: AstrMessageEvent) -> bool:
+    async def should_process_tts_request(event: AstrMessageEvent) -> bool:
         """检查是否应该处理TTS请求
 
         Args:
@@ -139,14 +151,14 @@ class SessionServiceManager:
 
         """
         session_id = event.unified_msg_origin
-        return SessionServiceManager.is_tts_enabled_for_session(session_id)
+        return await SessionServiceManager.is_tts_enabled_for_session(session_id)
 
     # =============================================================================
     # 会话整体启停相关方法
     # =============================================================================
 
     @staticmethod
-    def is_session_enabled(session_id: str) -> bool:
+    async def is_session_enabled(session_id: str) -> bool:
         """检查会话是否整体启用
 
         Args:
@@ -157,11 +169,11 @@ class SessionServiceManager:
 
         """
         # 获取会话服务配置
-        session_services = sp.get(
-            "session_service_config",
-            {},
+        session_services = await sp.get_async(
             scope="umo",
             scope_id=session_id,
+            key="session_service_config",
+            default={},
         )
 
         # 如果配置了该会话的整体状态，返回该状态