diff --git a/astrbot/core/agent/response.py b/astrbot/core/agent/response.py index 3f3430c87..9e61fa8c7 100644 --- a/astrbot/core/agent/response.py +++ b/astrbot/core/agent/response.py @@ -1,7 +1,8 @@ import typing as T -from dataclasses import dataclass +from dataclasses import dataclass, field from astrbot.core.message.message_event_result import MessageChain +from astrbot.core.provider.entities import TokenUsage class AgentResponseData(T.TypedDict): @@ -12,3 +13,23 @@ class AgentResponseData(T.TypedDict): class AgentResponse: type: str data: AgentResponseData + + +@dataclass +class AgentStats: + token_usage: TokenUsage = field(default_factory=TokenUsage) + start_time: float = 0.0 + end_time: float = 0.0 + time_to_first_token: float = 0.0 + + @property + def duration(self) -> float: + return self.end_time - self.start_time + + def to_dict(self) -> dict: + return { + "token_usage": self.token_usage.__dict__, + "start_time": self.start_time, + "end_time": self.end_time, + "time_to_first_token": self.time_to_first_token, + } diff --git a/astrbot/core/agent/runners/tool_loop_agent_runner.py b/astrbot/core/agent/runners/tool_loop_agent_runner.py index 450e4dbcb..069de144f 100644 --- a/astrbot/core/agent/runners/tool_loop_agent_runner.py +++ b/astrbot/core/agent/runners/tool_loop_agent_runner.py @@ -1,4 +1,5 @@ import sys +import time import traceback import typing as T @@ -12,6 +13,7 @@ from mcp.types import ( ) from astrbot import logger +from astrbot.core.message.components import Json from astrbot.core.message.message_event_result import ( MessageChain, ) @@ -24,7 +26,7 @@ from astrbot.core.provider.provider import Provider from ..hooks import BaseAgentRunHooks from ..message import AssistantMessageSegment, Message, ToolCallMessageSegment -from ..response import AgentResponseData +from ..response import AgentResponseData, AgentStats from ..run_context import ContextWrapper, TContext from ..tool_executor import BaseFunctionToolExecutor from .base import AgentResponse, AgentState, BaseAgentRunner @@ -69,6 +71,9 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]): ) self.run_context.messages = messages + self.stats = AgentStats() + self.stats.start_time = time.time() + async def _iter_llm_responses(self) -> T.AsyncGenerator[LLMResponse, None]: """Yields chunks *and* a final LLMResponse.""" if self.streaming: @@ -98,6 +103,10 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]): async for llm_response in self._iter_llm_responses(): if llm_response.is_chunk: + # update ttft + if self.stats.time_to_first_token == 0: + self.stats.time_to_first_token = time.time() - self.stats.start_time + if llm_response.result_chain: yield AgentResponse( type="streaming_delta", @@ -121,6 +130,10 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]): ) continue llm_resp_result = llm_response + + if not llm_response.is_chunk and llm_response.usage: + # only count the token usage of the final response for computation purpose + self.stats.token_usage += llm_response.usage break # got final response if not llm_resp_result: @@ -132,6 +145,7 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]): if llm_resp.role == "err": # 如果 LLM 响应错误，转换到错误状态 self.final_llm_resp = llm_resp + self.stats.end_time = time.time() self._transition_state(AgentState.ERROR) yield AgentResponse( type="err", @@ -146,6 +160,7 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]): # 如果没有工具调用，转换到完成状态 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( @@ -175,23 +190,19 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]): # 如果有工具调用，还需处理工具调用 if llm_resp.tools_call_name: tool_call_result_blocks = [] - for tool_call_name in llm_resp.tools_call_name: - yield AgentResponse( - type="tool_call", - data=AgentResponseData( - chain=MessageChain(type="tool_call").message( - f"🔨 调用工具: {tool_call_name}" - ), - ), - ) async for result in self._handle_function_tools(self.req, llm_resp): if isinstance(result, list): tool_call_result_blocks = result elif isinstance(result, MessageChain): if result.type is None: - result.type = "tool_call_result" + # should not happen + continue + if result.type == "tool_direct_result": + ar_type = "tool_call_result" + else: + ar_type = result.type yield AgentResponse( - type="tool_call_result", + type=ar_type, data=AgentResponseData(chain=result), ) # 将结果添加到上下文中 @@ -234,6 +245,19 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]): llm_response.tools_call_args, llm_response.tools_call_ids, ): + yield MessageChain( + type="tool_call", + chain=[ + Json( + data={ + "id": func_tool_id, + "name": func_tool_name, + "args": func_tool_args, + "ts": time.time(), + } + ) + ], + ) try: if not req.func_tool: return @@ -307,7 +331,6 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]): content=res.content[0].text, ), ) - yield MessageChain().message(res.content[0].text) elif isinstance(res.content[0], ImageContent): tool_call_result_blocks.append( ToolCallMessageSegment( @@ -329,7 +352,6 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]): content=resource.text, ), ) - yield MessageChain().message(resource.text) elif ( isinstance(resource, BlobResourceContents) and resource.mimeType @@ -353,7 +375,22 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]): content="返回的数据类型不受支持", ), ) - yield MessageChain().message("返回的数据类型不受支持。") + + # yield the last tool call result + if tool_call_result_blocks: + last_tcr_content = str(tool_call_result_blocks[-1].content) + yield MessageChain( + type="tool_call_result", + chain=[ + Json( + data={ + "id": func_tool_id, + "ts": time.time(), + "result": last_tcr_content, + } + ) + ], + ) elif resp is None: # Tool 直接请求发送消息给用户 @@ -363,6 +400,7 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]): f"{func_tool_name} 没有没有返回值或者将结果直接发送给用户，此工具调用不会被记录到历史中。" ) self._transition_state(AgentState.DONE) + self.stats.end_time = time.time() else: # 不应该出现其他类型 logger.warning( diff --git a/astrbot/core/astr_agent_run_util.py b/astrbot/core/astr_agent_run_util.py index d94d96a82..5421a14c0 100644 --- a/astrbot/core/astr_agent_run_util.py +++ b/astrbot/core/astr_agent_run_util.py @@ -4,6 +4,7 @@ from collections.abc import AsyncGenerator from astrbot.core import logger from astrbot.core.agent.runners.tool_loop_agent_runner import ToolLoopAgentRunner from astrbot.core.astr_agent_context import AstrAgentContext +from astrbot.core.message.components import Json from astrbot.core.message.message_event_result import ( MessageChain, MessageEventResult, @@ -33,16 +34,27 @@ async def run_agent( msg_chain = resp.data["chain"] if msg_chain.type == "tool_direct_result": # tool_direct_result 用于标记 llm tool 需要直接发送给用户的内容 - await astr_event.send(resp.data["chain"]) + await astr_event.send(msg_chain) continue + if astr_event.get_platform_id() == "webchat": + await astr_event.send(msg_chain) # 对于其他情况，暂时先不处理 continue elif resp.type == "tool_call": if agent_runner.streaming: # 用来标记流式响应需要分节 yield MessageChain(chain=[], type="break") - if show_tool_use: + + if astr_event.get_platform_name() == "webchat": await astr_event.send(resp.data["chain"]) + elif show_tool_use: + json_comp = resp.data["chain"].chain[0] + if isinstance(json_comp, Json): + m = f"🔨 调用工具: {json_comp.data.get('name')}" + else: + m = "🔨 调用工具..." + chain = MessageChain(type="tool_call").message(m) + await astr_event.send(chain) continue if stream_to_general and resp.type == "streaming_delta": @@ -69,6 +81,15 @@ async def run_agent( continue yield resp.data["chain"] # MessageChain if agent_runner.done(): + # send agent stats to webchat + if astr_event.get_platform_name() == "webchat": + await astr_event.send( + MessageChain( + type="agent_stats", + chain=[Json(data=agent_runner.stats.to_dict())], + ) + ) + break except Exception as e: diff --git a/astrbot/core/message/components.py b/astrbot/core/message/components.py index 0e7b3bab6..050e36521 100644 --- a/astrbot/core/message/components.py +++ b/astrbot/core/message/components.py @@ -629,12 +629,11 @@ class Nodes(BaseMessageComponent): class Json(BaseMessageComponent): type = ComponentType.Json - data: str | dict - resid: int | None = 0 + data: dict - def __init__(self, data, **_): - if isinstance(data, dict): - data = json.dumps(data) + def __init__(self, data: str | dict, **_): + if isinstance(data, str): + data = json.loads(data) super().__init__(data=data, **_) diff --git a/astrbot/core/platform/sources/telegram/tg_event.py b/astrbot/core/platform/sources/telegram/tg_event.py index 37f60e65a..5faba6803 100644 --- a/astrbot/core/platform/sources/telegram/tg_event.py +++ b/astrbot/core/platform/sources/telegram/tg_event.py @@ -200,6 +200,15 @@ class TelegramPlatformEvent(AstrMessageEvent): if isinstance(chain, MessageChain): if chain.type == "break": # 分割符 + if message_id: + try: + await self.client.edit_message_text( + text=delta, + chat_id=payload["chat_id"], + message_id=message_id, + ) + except Exception as e: + logger.warning(f"编辑消息失败(streaming-break): {e!s}") message_id = None # 重置消息 ID delta = "" # 重置 delta continue diff --git a/astrbot/core/platform/sources/webchat/webchat_event.py b/astrbot/core/platform/sources/webchat/webchat_event.py index 9f1a6d059..2e529bb1d 100644 --- a/astrbot/core/platform/sources/webchat/webchat_event.py +++ b/astrbot/core/platform/sources/webchat/webchat_event.py @@ -1,11 +1,12 @@ import base64 +import json import os import shutil import uuid from astrbot.api import logger from astrbot.api.event import AstrMessageEvent, MessageChain -from astrbot.api.message_components import File, Image, Plain, Record +from astrbot.api.message_components import File, Image, Json, Plain, Record from astrbot.core.utils.astrbot_path import get_astrbot_data_path from .webchat_queue_mgr import webchat_queue_mgr @@ -41,12 +42,20 @@ class WebChatMessageEvent(AstrMessageEvent): await web_chat_back_queue.put( { "type": "plain", - "cid": cid, "data": data, "streaming": streaming, "chain_type": message.type, }, ) + elif isinstance(comp, Json): + await web_chat_back_queue.put( + { + "type": "plain", + "data": json.dumps(comp.data, ensure_ascii=False), + "streaming": streaming, + "chain_type": message.type, + }, + ) elif isinstance(comp, Image): # save image to local filename = f"{str(uuid.uuid4())}.jpg" @@ -58,7 +67,6 @@ class WebChatMessageEvent(AstrMessageEvent): await web_chat_back_queue.put( { "type": "image", - "cid": cid, "data": data, "streaming": streaming, }, @@ -74,7 +82,6 @@ class WebChatMessageEvent(AstrMessageEvent): await web_chat_back_queue.put( { "type": "record", - "cid": cid, "data": data, "streaming": streaming, }, @@ -91,7 +98,6 @@ class WebChatMessageEvent(AstrMessageEvent): await web_chat_back_queue.put( { "type": "file", - "cid": cid, "data": data, "streaming": streaming, }, @@ -111,18 +117,17 @@ class WebChatMessageEvent(AstrMessageEvent): cid = self.session_id.split("!")[-1] web_chat_back_queue = webchat_queue_mgr.get_or_create_back_queue(cid) async for chain in generator: - if chain.type == "break" and final_data: - # 分割符 - await web_chat_back_queue.put( - { - "type": "break", # break means a segment end - "data": final_data, - "streaming": True, - "cid": cid, - }, - ) - final_data = "" - continue + # if chain.type == "break" and final_data: + # # 分割符 + # await web_chat_back_queue.put( + # { + # "type": "break", # break means a segment end + # "data": final_data, + # "streaming": True, + # }, + # ) + # final_data = "" + # continue r = await WebChatMessageEvent._send( chain, @@ -142,7 +147,6 @@ class WebChatMessageEvent(AstrMessageEvent): "data": final_data, "reasoning": reasoning_content, "streaming": True, - "cid": cid, }, ) await super().send_streaming(generator, use_fallback) diff --git a/astrbot/core/provider/entities.py b/astrbot/core/provider/entities.py index dc188f141..d13e9b56a 100644 --- a/astrbot/core/provider/entities.py +++ b/astrbot/core/provider/entities.py @@ -1,3 +1,5 @@ +from __future__ import annotations + import base64 import enum import json @@ -199,6 +201,38 @@ class ProviderRequest: return "" +@dataclass +class TokenUsage: + input_other: int = 0 + """The number of input tokens, excluding cached tokens.""" + input_cached: int = 0 + """The number of input cached tokens.""" + output: int = 0 + """The number of output tokens.""" + + @property + def total(self) -> int: + return self.input_other + self.input_cached + self.output + + @property + def input(self) -> int: + return self.input_other + self.input_cached + + def __add__(self, other: TokenUsage) -> TokenUsage: + return TokenUsage( + input_other=self.input_other + other.input_other, + input_cached=self.input_cached + other.input_cached, + output=self.output + other.output, + ) + + def __sub__(self, other: TokenUsage) -> TokenUsage: + return TokenUsage( + input_other=self.input_other - other.input_other, + input_cached=self.input_cached - other.input_cached, + output=self.output - other.output, + ) + + @dataclass class LLMResponse: role: str @@ -227,6 +261,11 @@ class LLMResponse: is_chunk: bool = False """Indicates if the response is a chunked response.""" + id: str | None = None + """The ID of the response. For chunked responses, it's the ID of the chunk; for non-chunked responses, it's the ID of the response.""" + usage: TokenUsage | None = None + """The usage of the response. For chunked responses, it's the usage of the chunk; for non-chunked responses, it's the usage of the response.""" + def __init__( self, role: str, @@ -241,6 +280,8 @@ class LLMResponse: | AnthropicMessage | None = None, is_chunk: bool = False, + id: str | None = None, + usage: TokenUsage | None = None, ): """初始化 LLMResponse diff --git a/astrbot/core/provider/sources/anthropic_source.py b/astrbot/core/provider/sources/anthropic_source.py index bd0f06fba..7e33f40d9 100644 --- a/astrbot/core/provider/sources/anthropic_source.py +++ b/astrbot/core/provider/sources/anthropic_source.py @@ -6,10 +6,12 @@ from mimetypes import guess_type import anthropic from anthropic import AsyncAnthropic from anthropic.types import Message +from anthropic.types.message_delta_usage import MessageDeltaUsage +from anthropic.types.usage import Usage from astrbot import logger from astrbot.api.provider import Provider -from astrbot.core.provider.entities import LLMResponse +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 @@ -107,6 +109,22 @@ class ProviderAnthropic(Provider): return system_prompt, new_messages + def _extract_usage(self, usage: Usage) -> TokenUsage: + # https://docs.claude.com/en/docs/build-with-claude/prompt-caching#tracking-cache-performance + return TokenUsage( + input_other=usage.input_tokens or 0, + input_cached=usage.cache_read_input_tokens or 0, + output=usage.output_tokens, + ) + + def _update_usage(self, token_usage: TokenUsage, usage: MessageDeltaUsage) -> None: + if usage.input_tokens is not None: + token_usage.input_other = usage.input_tokens + if usage.cache_read_input_tokens is not None: + token_usage.input_cached = usage.cache_read_input_tokens + if usage.output_tokens is not None: + token_usage.output = usage.output_tokens + async def _query(self, payloads: dict, tools: ToolSet | None) -> LLMResponse: if tools: if tool_list := tools.get_func_desc_anthropic_style(): @@ -131,6 +149,10 @@ class ProviderAnthropic(Provider): llm_response.tools_call_args.append(content_block.input) llm_response.tools_call_name.append(content_block.name) llm_response.tools_call_ids.append(content_block.id) + + llm_response.id = completion.id + llm_response.usage = self._extract_usage(completion.usage) + # TODO(Soulter): 处理 end_turn 情况 if not llm_response.completion_text and not llm_response.tools_call_args: raise Exception(f"Anthropic API 返回的 completion 无法解析：{completion}。") @@ -152,9 +174,16 @@ class ProviderAnthropic(Provider): final_text = "" final_tool_calls = [] + id = None + usage = TokenUsage() + async with self.client.messages.stream(**payloads) as stream: assert isinstance(stream, anthropic.AsyncMessageStream) async for event in stream: + if event.type == "message_start": + # the usage contains input token usage + id = event.message.id + usage = self._extract_usage(event.message.usage) if event.type == "content_block_start": if event.content_block.type == "text": # 文本块开始 @@ -162,6 +191,8 @@ class ProviderAnthropic(Provider): role="assistant", completion_text="", is_chunk=True, + usage=usage, + id=id, ) elif event.content_block.type == "tool_use": # 工具使用块开始，初始化缓冲区 @@ -179,6 +210,8 @@ class ProviderAnthropic(Provider): role="assistant", completion_text=event.delta.text, is_chunk=True, + usage=usage, + id=id, ) elif event.delta.type == "input_json_delta": # 工具调用参数增量 @@ -215,6 +248,8 @@ class ProviderAnthropic(Provider): tools_call_name=[tool_info["name"]], tools_call_ids=[tool_info["id"]], is_chunk=True, + usage=usage, + id=id, ) except json.JSONDecodeError: # JSON 解析失败，跳过这个工具调用 @@ -223,11 +258,17 @@ class ProviderAnthropic(Provider): # 清理缓冲区 del tool_use_buffer[event.index] + elif event.type == "message_delta": + if event.usage: + self._update_usage(usage, event.usage) + # 返回最终的完整结果 final_response = LLMResponse( role="assistant", completion_text=final_text, is_chunk=False, + usage=usage, + id=id, ) if final_tool_calls: diff --git a/astrbot/core/provider/sources/gemini_source.py b/astrbot/core/provider/sources/gemini_source.py index e2efc6aab..8e0b89081 100644 --- a/astrbot/core/provider/sources/gemini_source.py +++ b/astrbot/core/provider/sources/gemini_source.py @@ -14,7 +14,7 @@ import astrbot.core.message.components as Comp from astrbot import logger from astrbot.api.provider import Provider from astrbot.core.message.message_event_result import MessageChain -from astrbot.core.provider.entities import LLMResponse +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 @@ -347,6 +347,16 @@ class ProviderGoogleGenAI(Provider): ] return "".join(thought_buf).strip() + def _extract_usage( + self, usage_metadata: types.GenerateContentResponseUsageMetadata + ) -> TokenUsage: + """Extract usage from candidate""" + return TokenUsage( + input_other=usage_metadata.prompt_token_count or 0, + input_cached=usage_metadata.cached_content_token_count or 0, + output=usage_metadata.candidates_token_count or 0, + ) + def _process_content_parts( self, candidate: types.Candidate, @@ -501,6 +511,9 @@ class ProviderGoogleGenAI(Provider): result.candidates[0], llm_response, ) + llm_response.id = result.response_id + if result.usage_metadata: + llm_response.usage = self._extract_usage(result.usage_metadata) return llm_response async def _query_stream( @@ -569,6 +582,9 @@ class ProviderGoogleGenAI(Provider): chunk.candidates[0], llm_response, ) + llm_response.id = chunk.response_id + if chunk.usage_metadata: + llm_response.usage = self._extract_usage(chunk.usage_metadata) yield llm_response return @@ -596,6 +612,9 @@ class ProviderGoogleGenAI(Provider): chunk.candidates[0], final_response, ) + final_response.id = chunk.response_id + if chunk.usage_metadata: + final_response.usage = self._extract_usage(chunk.usage_metadata) break # Yield final complete response with accumulated text diff --git a/astrbot/core/provider/sources/openai_source.py b/astrbot/core/provider/sources/openai_source.py index 788b649a9..4aeacf672 100644 --- a/astrbot/core/provider/sources/openai_source.py +++ b/astrbot/core/provider/sources/openai_source.py @@ -12,6 +12,7 @@ from openai._exceptions import NotFoundError from openai.lib.streaming.chat._completions import ChatCompletionStreamState from openai.types.chat.chat_completion import ChatCompletion from openai.types.chat.chat_completion_chunk import ChatCompletionChunk +from openai.types.completion_usage import CompletionUsage import astrbot.core.message.components as Comp from astrbot import logger @@ -19,7 +20,7 @@ from astrbot.api.provider import Provider from astrbot.core.agent.message import Message from astrbot.core.agent.tool import ToolSet from astrbot.core.message.message_event_result import MessageChain -from astrbot.core.provider.entities import LLMResponse, ToolCallsResult +from astrbot.core.provider.entities import LLMResponse, TokenUsage, ToolCallsResult from astrbot.core.utils.io import download_image_by_url from ..register import register_provider_adapter @@ -208,6 +209,7 @@ class ProviderOpenAIOfficial(Provider): # handle the content delta reasoning = self._extract_reasoning_content(chunk) _y = False + llm_response.id = chunk.id if reasoning: llm_response.reasoning_content = reasoning _y = True @@ -217,6 +219,8 @@ class ProviderOpenAIOfficial(Provider): chain=[Comp.Plain(completion_text)], ) _y = True + if chunk.usage: + llm_response.usage = self._extract_usage(chunk.usage) if _y: yield llm_response @@ -245,6 +249,15 @@ class ProviderOpenAIOfficial(Provider): reasoning_text = str(reasoning_attr) return reasoning_text + def _extract_usage(self, usage: CompletionUsage) -> TokenUsage: + ptd = usage.prompt_tokens_details + cached = ptd.cached_tokens if ptd and ptd.cached_tokens else 0 + 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, + ) + async def _parse_openai_completion( self, completion: ChatCompletion, tools: ToolSet | None ) -> LLMResponse: @@ -321,6 +334,10 @@ class ProviderOpenAIOfficial(Provider): raise Exception(f"API 返回的 completion 无法解析：{completion}。") llm_response.raw_completion = completion + llm_response.id = completion.id + + if completion.usage: + llm_response.usage = self._extract_usage(completion.usage) return llm_response diff --git a/astrbot/core/star/context.py b/astrbot/core/star/context.py index 9a52ec8bc..2561762f1 100644 --- a/astrbot/core/star/context.py +++ b/astrbot/core/star/context.py @@ -296,6 +296,10 @@ class Context: provider_type=ProviderType.CHAT_COMPLETION, umo=umo, ) + if prov is None: + raise ProviderNotFoundError( + "provider not found, please choose provider first" + ) if not isinstance(prov, Provider): raise ValueError("返回的 Provider 不是 Provider 类型") return prov diff --git a/astrbot/dashboard/routes/chat.py b/astrbot/dashboard/routes/chat.py index f2439c058..c2b991ef7 100644 --- a/astrbot/dashboard/routes/chat.py +++ b/astrbot/dashboard/routes/chat.py @@ -227,16 +227,19 @@ class ChatRoute(Route): text: str, media_parts: list, reasoning: str, + agent_stats: dict, ): """保存 bot 消息到历史记录，返回保存的记录""" bot_message_parts = [] + bot_message_parts.extend(media_parts) if text: bot_message_parts.append({"type": "plain", "text": text}) - bot_message_parts.extend(media_parts) new_his = {"type": "bot", "message": bot_message_parts} if reasoning: new_his["reasoning"] = reasoning + if agent_stats: + new_his["agent_stats"] = agent_stats record = await self.platform_history_mgr.insert( platform_id="webchat", @@ -294,7 +297,8 @@ class ChatRoute(Route): accumulated_parts = [] accumulated_text = "" accumulated_reasoning = "" - + tool_calls = {} + agent_stats = {} try: async with track_conversation(self.running_convs, webchat_conv_id): while True: @@ -314,6 +318,16 @@ class ChatRoute(Route): result_text = result["data"] msg_type = result.get("type") streaming = result.get("streaming", False) + chain_type = result.get("chain_type") + + if chain_type == "agent_stats": + stats_info = { + "type": "agent_stats", + "data": json.loads(result_text), + } + yield f"data: {json.dumps(stats_info, ensure_ascii=False)}\n\n" + agent_stats = stats_info["data"] + continue # 发送 SSE 数据 try: @@ -335,8 +349,30 @@ class ChatRoute(Route): # 累积消息部分 if msg_type == "plain": - chain_type = result.get("chain_type", "normal") - if chain_type == "reasoning": + chain_type = result.get("chain_type") + if chain_type == "tool_call": + tool_call = json.loads(result_text) + tool_calls[tool_call.get("id")] = tool_call + if accumulated_text: + # 如果累积了文本，则先保存文本 + accumulated_parts.append( + {"type": "plain", "text": accumulated_text} + ) + accumulated_text = "" + elif chain_type == "tool_call_result": + tcr = json.loads(result_text) + tc_id = tcr.get("id") + if tc_id in tool_calls: + tool_calls[tc_id]["result"] = tcr.get("result") + tool_calls[tc_id]["finished_ts"] = tcr.get("ts") + accumulated_parts.append( + { + "type": "tool_call", + "tool_calls": [tool_calls[tc_id]], + } + ) + tool_calls.pop(tc_id, None) + elif chain_type == "reasoning": accumulated_reasoning += result_text elif streaming: accumulated_text += result_text @@ -369,15 +405,20 @@ class ChatRoute(Route): if msg_type == "end": break elif ( - (streaming and msg_type == "complete") - or not streaming - or msg_type == "break" + (streaming and msg_type == "complete") or not streaming + # or msg_type == "break" ): + if ( + chain_type == "tool_call" + or chain_type == "tool_call_result" + ): + continue saved_record = await self._save_bot_message( webchat_conv_id, accumulated_text, accumulated_parts, accumulated_reasoning, + agent_stats, ) # 发送保存的消息信息给前端 if saved_record and not client_disconnected: @@ -392,11 +433,11 @@ class ChatRoute(Route): yield f"data: {json.dumps(saved_info, ensure_ascii=False)}\n\n" except Exception: pass - # 重置累积变量 (对于 break 后的下一段消息) - if msg_type == "break": - accumulated_parts = [] - accumulated_text = "" - accumulated_reasoning = "" + accumulated_parts = [] + accumulated_text = "" + accumulated_reasoning = "" + tool_calls = {} + agent_stats = {} except BaseException as e: logger.exception(f"WebChat stream unexpected error: {e}", exc_info=True) diff --git a/dashboard/src/components/chat/Chat.vue b/dashboard/src/components/chat/Chat.vue index 509971ca8..5524e787d 100644 --- a/dashboard/src/components/chat/Chat.vue +++ b/dashboard/src/components/chat/Chat.vue @@ -575,5 +575,9 @@ onBeforeUnmount(() => { .chat-page-container { padding: 0 !important; } + + .conversation-header { + padding: 2px; + } } diff --git a/dashboard/src/components/chat/MessageList.vue b/dashboard/src/components/chat/MessageList.vue index 8361b5176..cd14c6574 100644 --- a/dashboard/src/components/chat/MessageList.vue +++ b/dashboard/src/components/chat/MessageList.vue @@ -5,56 +5,66 @@
-
- -
- mdi-reply - {{ getReplyContent(msg.content.reply_to.message_id) }} -
- 
{{ msg.content.message }}
- - -
-
- + +
- - - mdi-star-four-points-small + + mdi-star-four-points-small
@@ -62,10 +72,11 @@
{{ tm('message.loading') }}
- +
- {{ formatMessageTime(msg.created_at) }} + {{ formatMessageTime(msg.created_at) + }} + + + + + +
+ {{ tm('stats.inputTokens') }} + {{ getInputTokens(msg.content.agentStats.token_usage) }} +
+
+ {{ tm('stats.outputTokens') }} + {{ msg.content.agentStats.token_usage.output || 0 }} +
+
+ {{ tm('stats.cachedTokens') }} + {{ msg.content.agentStats.token_usage.input_cached }} +
+
+ {{ tm('stats.ttft') }} + {{ formatTTFT(msg.content.agentStats.time_to_first_token) }} +
+
+ {{ tm('stats.duration') }} + {{ formatAgentDuration(msg.content.agentStats) }} +
+ + + @@ -191,6 +287,9 @@ export default { scrollTimer: null, expandedReasoning: new Set(), // Track which reasoning blocks are expanded downloadingFiles: new Set(), // Track which files are being downloaded + expandedToolCalls: new Set(), // Track which tool call cards are expanded + elapsedTimeTimer: null, // Timer for updating elapsed time + currentTime: Date.now() / 1000, // Current time for elapsed time calculation }; }, mounted() { @@ -198,6 +297,7 @@ export default { this.initImageClickEvents(); this.addScrollListener(); this.scrollToBottom(); + this.startElapsedTimeTimer(); }, updated() { this.initCodeCopyButtons(); @@ -207,6 +307,12 @@ export default { } }, methods: { + // 检查 message 中是否有音频 + hasAudio(messageParts) { + if (!Array.isArray(messageParts)) return false; + return messageParts.some(part => part.type === 'record' && part.embedded_url); + }, + // 获取被引用消息的内容 getReplyContent(messageId) { const replyMsg = this.messages.find(m => m.id === messageId); @@ -214,9 +320,7 @@ export default { return this.tm('reply.notFound'); } let content = ''; - if (typeof replyMsg.content.message === 'string') { - content = replyMsg.content.message; - } else if (Array.isArray(replyMsg.content.message)) { + if (Array.isArray(replyMsg.content.message)) { const textParts = replyMsg.content.message .filter(part => part.type === 'plain' && part.text) .map(part => part.text); @@ -233,7 +337,7 @@ export default { scrollToMessage(messageId) { const msgIndex = this.messages.findIndex(m => m.id === messageId); if (msgIndex === -1) return; - + const container = this.$refs.messageContainer; const messageItems = container?.querySelectorAll('.message-item'); if (messageItems && messageItems[msgIndex]) { @@ -265,16 +369,16 @@ export default { // 下载文件 async downloadFile(file) { if (!file.attachment_id) return; - + // 标记为下载中 this.downloadingFiles.add(file.attachment_id); this.downloadingFiles = new Set(this.downloadingFiles); - + try { const response = await axios.get(`/api/chat/get_attachment?attachment_id=${file.attachment_id}`, { responseType: 'blob' }); - + const url = URL.createObjectURL(response.data); const a = document.createElement('a'); a.href = url; @@ -313,29 +417,29 @@ export default { }, // 复制bot消息到剪贴板 - copyBotMessage(message, messageIndex) { - // 获取对应的消息对象 - const msgObj = this.messages[messageIndex].content; + copyBotMessage(messageParts, messageIndex) { let textToCopy = ''; - // 如果有文本消息，添加到复制内容中 - if (message && message.trim()) { - // 移除HTML标签，获取纯文本 - const tempDiv = document.createElement('div'); - tempDiv.innerHTML = message; - textToCopy = tempDiv.textContent || tempDiv.innerText || message; - } + if (Array.isArray(messageParts)) { + // 提取所有文本内容 + const textContents = messageParts + .filter(part => part.type === 'plain' && part.text) + .map(part => part.text); + textToCopy = textContents.join('\n'); - // 如果有内嵌图片，添加说明 - if (msgObj && msgObj.embedded_images && msgObj.embedded_images.length > 0) { - if (textToCopy) textToCopy += '\n\n'; - textToCopy += `[包含 ${msgObj.embedded_images.length} 张图片]`; - } + // 检查是否有图片 + const imageCount = messageParts.filter(part => part.type === 'image' && part.embedded_url).length; + if (imageCount > 0) { + if (textToCopy) textToCopy += '\n\n'; + textToCopy += `[包含 ${imageCount} 张图片]`; + } - // 如果有内嵌音频，添加说明 - if (msgObj && msgObj.embedded_audio) { - if (textToCopy) textToCopy += '\n\n'; - textToCopy += '[包含音频内容]'; + // 检查是否有音频 + const hasAudio = messageParts.some(part => part.type === 'record' && part.embedded_url); + if (hasAudio) { + if (textToCopy) textToCopy += '\n\n'; + textToCopy += '[包含音频内容]'; + } } // 如果没有任何内容，使用默认文本 @@ -487,26 +591,31 @@ export default { clearTimeout(this.scrollTimer); this.scrollTimer = null; } + // 清理 elapsed time 计时器 + if (this.elapsedTimeTimer) { + clearInterval(this.elapsedTimeTimer); + this.elapsedTimeTimer = null; + } }, // 格式化消息时间，支持别名显示 formatMessageTime(dateStr) { if (!dateStr) return ''; - + const date = new Date(dateStr); const now = new Date(); - + // 获取本地时间的日期部分 const dateDay = new Date(date.getFullYear(), date.getMonth(), date.getDate()); const todayDay = new Date(now.getFullYear(), now.getMonth(), now.getDate()); const yesterdayDay = new Date(todayDay); yesterdayDay.setDate(yesterdayDay.getDate() - 1); - + // 格式化时间 HH:MM const hours = date.getHours().toString().padStart(2, '0'); const minutes = date.getMinutes().toString().padStart(2, '0'); const timeStr = `${hours}:${minutes}`; - + // 判断是今天、昨天还是更早 if (dateDay.getTime() === todayDay.getTime()) { return `${this.tm('time.today')} ${timeStr}`; @@ -518,6 +627,114 @@ export default { const day = date.getDate().toString().padStart(2, '0'); return `${month}-${day} ${timeStr}`; } + }, + + // Tool call related methods + toggleToolCall(messageIndex, partIndex, toolCallIndex) { + const key = `${messageIndex}-${partIndex}-${toolCallIndex}`; + if (this.expandedToolCalls.has(key)) { + this.expandedToolCalls.delete(key); + } else { + this.expandedToolCalls.add(key); + } + // Force reactivity + this.expandedToolCalls = new Set(this.expandedToolCalls); + }, + + isToolCallExpanded(messageIndex, partIndex, toolCallIndex) { + return this.expandedToolCalls.has(`${messageIndex}-${partIndex}-${toolCallIndex}`); + }, + + // Start timer for updating elapsed time + startElapsedTimeTimer() { + // Update every 12ms for sub-second precision, then every second after 1s + let fastUpdateCount = 0; + const fastUpdateInterval = 12; + const slowUpdateInterval = 1000; + + const updateTime = () => { + this.currentTime = Date.now() / 1000; + + // Check if there are any running tool calls + const hasRunningToolCalls = this.messages.some(msg => + Array.isArray(msg.content.message) && msg.content.message.some(part => + part.type === 'tool_call' && part.tool_calls?.some(tc => !tc.finished_ts) + ) + ); + + if (hasRunningToolCalls) { + // Check if any running tool call is under 1 second + const hasSubSecondToolCall = this.messages.some(msg => + Array.isArray(msg.content.message) && msg.content.message.some(part => + part.type === 'tool_call' && part.tool_calls?.some(tc => + !tc.finished_ts && (this.currentTime - tc.ts) < 1 + ) + ) + ); + + if (hasSubSecondToolCall) { + fastUpdateCount++; + this.elapsedTimeTimer = setTimeout(updateTime, fastUpdateInterval); + } else { + this.elapsedTimeTimer = setTimeout(updateTime, slowUpdateInterval); + } + } else { + // No running tool calls, check again after 1 second + this.elapsedTimeTimer = setTimeout(updateTime, slowUpdateInterval); + } + }; + + updateTime(); + }, + + // Get elapsed time string for a tool call + getElapsedTime(startTs) { + const elapsed = this.currentTime - startTs; + return this.formatDuration(elapsed); + }, + + // Format duration in seconds to human readable string + formatDuration(seconds) { + if (seconds < 1) { + return `${Math.round(seconds * 1000)}ms`; + } else if (seconds < 60) { + return `${seconds.toFixed(1)}s`; + } else { + const minutes = Math.floor(seconds / 60); + const secs = Math.round(seconds % 60); + return `${minutes}m ${secs}s`; + } + }, + + // Format tool result for display + formatToolResult(result) { + if (!result) return ''; + // Try to parse as JSON for pretty formatting + try { + const parsed = JSON.parse(result); + return JSON.stringify(parsed, null, 2); + } catch { + return result; + } + }, + + // Get input tokens (input_other + input_cached) + getInputTokens(tokenUsage) { + if (!tokenUsage) return 0; + return (tokenUsage.input_other || 0) + (tokenUsage.input_cached || 0); + }, + + // Format agent duration + formatAgentDuration(agentStats) { + if (!agentStats) return ''; + const duration = agentStats.end_time - agentStats.start_time; + return this.formatDuration(duration); + }, + + // Format time to first token + formatTTFT(ttft) { + if (!ttft || ttft <= 0) return ''; + return this.formatDuration(ttft); } } } @@ -548,6 +765,22 @@ export default { min-height: 0; } +.message-bubble { + padding: 2px 16px; + border-radius: 12px; +} + + +@media (max-width: 768px) { + .messages-container { + padding: 0; + } + + .message-bubble { + padding: 2px 8px; + } +} + /* 消息列表样式 */ .message-list { max-width: 900px; @@ -603,6 +836,19 @@ export default { white-space: nowrap; } +/* Agent Stats Info Icon */ +.stats-info-icon { + margin-left: 6px; + color: var(--v-theme-secondaryText); + opacity: 0.6; + cursor: pointer; + transition: opacity 0.2s ease; +} + +.stats-info-icon:hover { + opacity: 1; +} + .bot-message:hover .message-actions { opacity: 1; } @@ -679,15 +925,12 @@ export default { 0% { background-color: rgba(103, 58, 183, 0.3); } + 100% { background-color: transparent; } } -.message-bubble { - padding: 2px 16px; - border-radius: 12px; -} .user-bubble { color: var(--v-theme-primaryText); @@ -911,6 +1154,175 @@ export default { .v-theme--dark .reasoning-text { opacity: 0.85; } + +/* Tool Call Card Styles */ +.tool-calls-container { + display: flex; + flex-direction: column; + gap: 8px; + margin-bottom: 12px; + margin-top: 6px; +} + +.tool-call-card { + border-radius: 8px; + overflow: hidden; + background-color: #eff3f6; + margin: 8px 0px; +} + +.v-theme--dark .tool-call-card { + background-color: rgba(40, 60, 100, 0.4); + border-color: rgba(100, 140, 200, 0.4); +} + +.tool-call-header { + display: flex; + align-items: center; + padding: 10px 12px; + cursor: pointer; + user-select: none; + transition: background-color 0.2s ease; + gap: 8px; +} + +.tool-call-header:hover { + background-color: rgba(169, 194, 219, 0.15); +} + +.v-theme--dark .tool-call-header:hover { + background-color: rgba(100, 150, 200, 0.2); +} + +.tool-call-expand-icon { + color: var(--v-theme-secondary); + transition: transform 0.2s ease; + flex-shrink: 0; +} + +.tool-call-icon { + color: var(--v-theme-secondary); + flex-shrink: 0; +} + +.tool-call-info { + display: flex; + flex-direction: column; + gap: 2px; + flex: 1; + min-width: 0; +} + +.tool-call-name { + font-size: 13px; + font-weight: 600; + color: var(--v-theme-secondary); +} + +.tool-call-id { + font-size: 11px; + color: var(--v-theme-secondaryText); + opacity: 0.7; + overflow: hidden; + text-overflow: ellipsis; + white-space: nowrap; +} + +.tool-call-status { + margin-left: 8px; + display: flex; + align-items: center; + gap: 4px; + font-size: 12px; + font-weight: 500; + flex-shrink: 0; +} + +.tool-call-status.status-running { + color: #ff9800; +} + +.tool-call-status.status-finished { + color: #4caf50; +} + +.tool-call-status .status-icon { + font-size: 14px; +} + +.tool-call-status .status-icon.spinning { + animation: spin 1s linear infinite; +} + +@keyframes spin { + from { + transform: rotate(0deg); + } + + to { + transform: rotate(360deg); + } +} + +.tool-call-details { + padding: 12px; + background-color: rgba(255, 255, 255, 0.5); + animation: fadeIn 0.2s ease-in-out; +} + +.v-theme--dark .tool-call-details { + border-top-color: rgba(100, 140, 200, 0.3); + background-color: rgba(30, 45, 70, 0.5); +} + +.tool-call-detail-row { + display: flex; + flex-direction: column; + margin-bottom: 8px; +} + +.tool-call-detail-row:last-child { + margin-bottom: 0; +} + +.detail-label { + font-size: 11px; + font-weight: 600; + color: var(--v-theme-secondaryText); + text-transform: uppercase; + letter-spacing: 0.5px; + margin-bottom: 4px; +} + +.detail-value { + font-size: 12px; + color: var(--v-theme-primaryText); + background-color: transparent; + padding: 4px 8px; + border-radius: 4px; + word-break: break-all; +} + +.detail-json { + font-family: 'Fira Code', 'Consolas', monospace; + white-space: pre-wrap; + max-height: 200px; + overflow-y: auto; + margin: 0; +} + +.detail-result { + max-height: 300px; + background-color: transparent; +} + +.v-theme--dark .detail-value { + background-color: transparent; +} + +.v-theme--dark .detail-result { + background-color: transparent; +} diff --git a/dashboard/src/composables/useMessages.ts b/dashboard/src/composables/useMessages.ts index d50f8558e..44b0e59a2 100644 --- a/dashboard/src/composables/useMessages.ts +++ b/dashboard/src/composables/useMessages.ts @@ -2,19 +2,29 @@ import { ref, reactive, type Ref } from 'vue'; import axios from 'axios'; import { useToast } from '@/utils/toast'; -// 新格式消息部分的类型定义 -export interface MessagePart { - type: 'plain' | 'image' | 'record' | 'file' | 'video' | 'reply'; - text?: string; // for plain - attachment_id?: string; // for image, record, file, video - filename?: string; // for file (filename from backend) - message_id?: number; // for reply (PlatformSessionHistoryMessage.id) +// 工具调用信息 +export interface ToolCall { + id: string; + name: string; + args: Record; + ts: number; // 开始时间戳 + result?: string; // 工具调用结果 + finished_ts?: number; // 完成时间戳 } -// 引用信息 -export interface ReplyInfo { - messageId: number; - messageContent: string; +// Token 使用统计 +export interface TokenUsage { + input_other: number; + input_cached: number; + output: number; +} + +// Agent 统计信息 +export interface AgentStats { + token_usage: TokenUsage; + start_time: number; + end_time: number; + time_to_first_token: number; } // 文件信息结构 @@ -24,24 +34,33 @@ export interface FileInfo { attachment_id?: string; // 用于按需下载 } -// 引用消息信息 -export interface ReplyTo { - message_id: number; - message_content?: string; // 被引用消息的内容（解析后填充） +// 消息部分的类型定义 +export interface MessagePart { + type: 'plain' | 'image' | 'record' | 'file' | 'video' | 'reply' | 'tool_call'; + text?: string; // for plain + attachment_id?: string; // for image, record, file, video + filename?: string; // for file (filename from backend) + message_id?: number; // for reply (PlatformSessionHistoryMessage.id) + tool_calls?: ToolCall[]; // for tool_call + // embedded fields - 加载后填充 + embedded_url?: string; // blob URL for image, record + embedded_file?: FileInfo; // for file (保留 attachment_id 用于按需下载) + reply_content?: string; // for reply - 被引用消息的内容 } +// 引用信息 (用于发送消息时) +export interface ReplyInfo { + messageId: number; + messageContent: string; +} + +// 简化的消息内容结构 export interface MessageContent { - type: string; - message: string | MessagePart[]; // 支持旧格式(string)和新格式(MessagePart[]) - reasoning?: string; - image_url?: string[]; - audio_url?: string; - file_url?: FileInfo[]; - embedded_images?: string[]; - embedded_audio?: string; - embedded_files?: FileInfo[]; - isLoading?: boolean; - reply_to?: ReplyTo; // 引用的消息 + type: string; // 'user' | 'bot' + message: MessagePart[]; // 消息部分列表 (保持顺序) + reasoning?: string; // reasoning content (for bot) + isLoading?: boolean; // loading state + agentStats?: AgentStats; // agent 统计信息 (for bot) } export interface Message { @@ -93,52 +112,64 @@ export function useMessages( } } - // 解析新格式消息为旧格式兼容的结构 (用于显示) + // 解析消息内容，填充 embedded 字段 (保持原始顺序) async function parseMessageContent(content: any): Promise { const message = content.message; - // 如果 message 是数组 (新格式) - if (Array.isArray(message)) { - let textParts: string[] = []; - let imageUrls: string[] = []; - let audioUrl: string | undefined; - let fileInfos: FileInfo[] = []; - let replyTo: ReplyTo | undefined; + // 如果 message 是字符串 (旧格式)，转换为数组格式 + if (typeof message === 'string') { + const parts: MessagePart[] = []; + let text = message; + // 处理旧格式的特殊标记 + if (text.startsWith('[IMAGE]')) { + const img = text.replace('[IMAGE]', ''); + const imageUrl = await getMediaFile(img); + parts.push({ + type: 'image', + embedded_url: imageUrl + }); + } else if (text.startsWith('[RECORD]')) { + const audio = text.replace('[RECORD]', ''); + const audioUrl = await getMediaFile(audio); + parts.push({ + type: 'record', + embedded_url: audioUrl + }); + } else if (text) { + parts.push({ + type: 'plain', + text: text + }); + } + + content.message = parts; + return; + } + + // 如果 message 是数组 (新格式)，遍历并填充 embedded 字段 + if (Array.isArray(message)) { for (const part of message as MessagePart[]) { - if (part.type === 'plain' && part.text) { - textParts.push(part.text); - } else if (part.type === 'image' && part.attachment_id) { - const url = await getAttachment(part.attachment_id); - if (url) imageUrls.push(url); + if (part.type === 'image' && part.attachment_id) { + part.embedded_url = await getAttachment(part.attachment_id); } else if (part.type === 'record' && part.attachment_id) { - audioUrl = await getAttachment(part.attachment_id); + part.embedded_url = await getAttachment(part.attachment_id); } else if (part.type === 'file' && part.attachment_id) { // file 类型不预加载，保留 attachment_id 以便点击时下载 - fileInfos.push({ + part.embedded_file = { attachment_id: part.attachment_id, filename: part.filename || 'file' - }); - } else if (part.type === 'reply' && part.message_id) { - replyTo = { message_id: part.message_id }; + }; } - // video 类型可以后续扩展 - } - - // 转换为旧格式兼容的结构 - content.message = textParts.join('\n'); - content.reply_to = replyTo; - if (content.type === 'user') { - content.image_url = imageUrls.length > 0 ? imageUrls : undefined; - content.audio_url = audioUrl; - content.file_url = fileInfos.length > 0 ? fileInfos : undefined; - } else { - content.embedded_images = imageUrls.length > 0 ? imageUrls : undefined; - content.embedded_audio = audioUrl; - content.embedded_files = fileInfos.length > 0 ? fileInfos : undefined; + // plain, reply, tool_call, video 保持原样 } } - // 如果 message 是字符串 (旧格式)，保持原有处理逻辑 + + // 处理 agent_stats (snake_case -> camelCase) + if (content.agent_stats) { + content.agentStats = content.agent_stats; + delete content.agent_stats; + } } async function getSessionMessages(sessionId: string, router: any) { @@ -161,46 +192,10 @@ export function useMessages( }, 3000); } - // 处理历史消息中的媒体文件 + // 处理历史消息 for (let i = 0; i < history.length; i++) { let content = history[i].content; - - // 首先尝试解析新格式消息 await parseMessageContent(content); - - // 以下是旧格式的兼容处理 (message 是字符串的情况) - if (typeof content.message === 'string') { - if (content.message?.startsWith('[IMAGE]')) { - let img = content.message.replace('[IMAGE]', ''); - const imageUrl = await getMediaFile(img); - if (!content.embedded_images) { - content.embedded_images = []; - } - content.embedded_images.push(imageUrl); - content.message = ''; - } - - if (content.message?.startsWith('[RECORD]')) { - let audio = content.message.replace('[RECORD]', ''); - const audioUrl = await getMediaFile(audio); - content.embedded_audio = audioUrl; - content.message = ''; - } - } - - // 旧格式中的 image_url 和 audio_url 字段处理 - if (content.image_url && content.image_url.length > 0) { - for (let j = 0; j < content.image_url.length; j++) { - // 检查是否已经是 blob URL (新格式解析后的结果) - if (!content.image_url[j].startsWith('blob:')) { - content.image_url[j] = await getMediaFile(content.image_url[j]); - } - } - } - - if (content.audio_url && !content.audio_url.startsWith('blob:')) { - content.audio_url = await getMediaFile(content.audio_url); - } } messages.value = history; @@ -217,47 +212,66 @@ export function useMessages( selectedModelName: string, replyTo: ReplyInfo | null = null ) { - // Create user message + // 构建用户消息的 message 部分 + const userMessageParts: MessagePart[] = []; + + // 添加引用消息段 + if (replyTo) { + userMessageParts.push({ + type: 'reply', + message_id: replyTo.messageId, + reply_content: replyTo.messageContent + }); + } + + // 添加纯文本消息段 + if (prompt) { + userMessageParts.push({ + type: 'plain', + text: prompt + }); + } + + // 添加文件消息段 + for (const f of stagedFiles) { + const partType = f.type === 'image' ? 'image' : + f.type === 'record' ? 'record' : 'file'; + + // 获取嵌入 URL + const embeddedUrl = await getAttachment(f.attachment_id); + + userMessageParts.push({ + type: partType as 'image' | 'record' | 'file', + attachment_id: f.attachment_id, + filename: f.original_name, + embedded_url: partType !== 'file' ? embeddedUrl : undefined, + embedded_file: partType === 'file' ? { + attachment_id: f.attachment_id, + filename: f.original_name + } : undefined + }); + } + + // 添加录音（如果有） + if (audioName) { + userMessageParts.push({ + type: 'record', + embedded_url: audioName // 录音使用本地 URL + }); + } + + // 创建用户消息 const userMessage: MessageContent = { type: 'user', - message: prompt, - image_url: [], - audio_url: undefined, - file_url: [], - reply_to: replyTo ? { message_id: replyTo.messageId } : undefined + message: userMessageParts }; - // 分离图片和文件 - const imageFiles = stagedFiles.filter(f => f.type === 'image'); - const nonImageFiles = stagedFiles.filter(f => f.type !== 'image'); - - // 使用 attachment_id 获取图片内容（避免 blob URL 被 revoke 后 404） - if (imageFiles.length > 0) { - const imageUrls = await Promise.all( - imageFiles.map(f => getAttachment(f.attachment_id)) - ); - userMessage.image_url = imageUrls.filter(url => url !== ''); - } - - // 使用 blob URL 作为音频预览（录音不走 attachment） - if (audioName) { - userMessage.audio_url = audioName; - } - - // 文件不预加载，只显示文件名和 attachment_id - if (nonImageFiles.length > 0) { - userMessage.file_url = nonImageFiles.map(f => ({ - filename: f.original_name, - attachment_id: f.attachment_id - })); - } - messages.value.push({ content: userMessage }); // 添加一个加载中的机器人消息占位符 - const loadingMessage = reactive({ + const loadingMessage = reactive({ type: 'bot', - message: '', + message: [], reasoning: '', isLoading: true }); @@ -272,12 +286,11 @@ export function useMessages( // 收集所有 attachment_id const files = stagedFiles.map(f => f.attachment_id); - // 构建 message 参数 - // 当 files 或 reply 存在时，message 是 list，否则是 str + // 构建发送给后端的 message 参数 let messageToSend: string | MessagePart[]; if (files.length > 0 || replyTo) { const parts: MessagePart[] = []; - + // 添加引用消息段 if (replyTo) { parts.push({ @@ -285,7 +298,7 @@ export function useMessages( message_id: replyTo.messageId }); } - + // 添加纯文本消息段 if (prompt) { parts.push({ @@ -293,17 +306,17 @@ export function useMessages( text: prompt }); } - + // 添加文件消息段 for (const f of stagedFiles) { - const partType = f.type === 'image' ? 'image' : - f.type === 'record' ? 'record' : 'file'; + const partType = f.type === 'image' ? 'image' : + f.type === 'record' ? 'record' : 'file'; parts.push({ type: partType as 'image' | 'record' | 'file', attachment_id: f.attachment_id }); } - + messageToSend = parts; } else { messageToSend = prompt; @@ -331,7 +344,7 @@ export function useMessages( const reader = response.body!.getReader(); const decoder = new TextDecoder(); let in_streaming = false; - let message_obj: any = null; + let message_obj: MessageContent | null = null; isStreaming.value = true; @@ -378,8 +391,10 @@ export function useMessages( const imageUrl = await getMediaFile(img); let bot_resp: MessageContent = { type: 'bot', - message: '', - embedded_images: [imageUrl] + message: [{ + type: 'image', + embedded_url: imageUrl + }] }; messages.value.push({ content: bot_resp }); } else if (chunk_json.type === 'record') { @@ -387,43 +402,122 @@ export function useMessages( const audioUrl = await getMediaFile(audio); let bot_resp: MessageContent = { type: 'bot', - message: '', - embedded_audio: audioUrl + message: [{ + type: 'record', + embedded_url: audioUrl + }] }; messages.value.push({ content: bot_resp }); } else if (chunk_json.type === 'file') { // 格式: [FILE]filename|original_name let fileData = chunk_json.data.replace('[FILE]', ''); - let [filename, originalName] = fileData.includes('|') - ? fileData.split('|', 2) + let [filename, originalName] = fileData.includes('|') + ? fileData.split('|', 2) : [fileData, fileData]; const fileUrl = await getMediaFile(filename); let bot_resp: MessageContent = { type: 'bot', - message: '', - embedded_files: [{ - url: fileUrl, - filename: originalName + message: [{ + type: 'file', + embedded_file: { + url: fileUrl, + filename: originalName + } }] }; messages.value.push({ content: bot_resp }); } else if (chunk_json.type === 'plain') { const chain_type = chunk_json.chain_type || 'normal'; - if (!in_streaming) { - message_obj = reactive({ - type: 'bot', - message: chain_type === 'reasoning' ? '' : chunk_json.data, - reasoning: chain_type === 'reasoning' ? chunk_json.data : '', - }); - messages.value.push({ content: message_obj }); - in_streaming = true; - } else { - if (chain_type === 'reasoning') { - // 使用 reactive 对象，直接修改属性会触发响应式更新 - message_obj.reasoning = (message_obj.reasoning || '') + chunk_json.data; + if (chain_type === 'tool_call') { + // 解析工具调用数据 + const toolCallData = JSON.parse(chunk_json.data); + const toolCall: ToolCall = { + id: toolCallData.id, + name: toolCallData.name, + args: toolCallData.args, + ts: toolCallData.ts + }; + + if (!in_streaming) { + message_obj = reactive({ + type: 'bot', + message: [{ + type: 'tool_call', + tool_calls: [toolCall] + }] + }); + messages.value.push({ content: message_obj }); + in_streaming = true; } else { - message_obj.message = (message_obj.message || '') + chunk_json.data; + // 找到最后一个 tool_call part 或创建新的 + const lastPart = message_obj!.message[message_obj!.message.length - 1]; + if (lastPart?.type === 'tool_call') { + // 检查是否已存在相同id的tool_call + const existingIndex = lastPart.tool_calls!.findIndex((tc: ToolCall) => tc.id === toolCall.id); + if (existingIndex === -1) { + lastPart.tool_calls!.push(toolCall); + } + } else { + // 添加新的 tool_call part + message_obj!.message.push({ + type: 'tool_call', + tool_calls: [toolCall] + }); + } + } + } else if (chain_type === 'tool_call_result') { + // 解析工具调用结果数据 + const resultData = JSON.parse(chunk_json.data); + + if (message_obj) { + // 遍历所有 tool_call parts 找到对应的 tool_call + for (const part of message_obj.message) { + if (part.type === 'tool_call' && part.tool_calls) { + const toolCall = part.tool_calls.find((tc: ToolCall) => tc.id === resultData.id); + if (toolCall) { + toolCall.result = resultData.result; + toolCall.finished_ts = resultData.ts; + break; + } + } + } + } + } else if (chain_type === 'reasoning') { + if (!in_streaming) { + message_obj = reactive({ + type: 'bot', + message: [], + reasoning: chunk_json.data + }); + messages.value.push({ content: message_obj }); + in_streaming = true; + } else { + message_obj!.reasoning = (message_obj!.reasoning || '') + chunk_json.data; + } + } else { + // normal text + if (!in_streaming) { + message_obj = reactive({ + type: 'bot', + message: [{ + type: 'plain', + text: chunk_json.data + }] + }); + messages.value.push({ content: message_obj }); + in_streaming = true; + } else { + // 找到最后一个 plain part 或创建新的 + const lastPart = message_obj!.message[message_obj!.message.length - 1]; + if (lastPart?.type === 'plain') { + lastPart.text = (lastPart.text || '') + chunk_json.data; + } else { + message_obj!.message.push({ + type: 'plain', + text: chunk_json.data + }); + } } } } else if (chunk_json.type === 'update_title') { @@ -435,6 +529,11 @@ export function useMessages( lastBotMsg.id = chunk_json.data.id; lastBotMsg.created_at = chunk_json.data.created_at; } + } else if (chunk_json.type === 'agent_stats') { + // 更新当前 bot 消息的 agent 统计信息 + if (message_obj) { + message_obj.agentStats = chunk_json.data; + } } if ((chunk_json.type === 'break' && chunk_json.streaming) || !chunk_json.streaming) { @@ -480,4 +579,3 @@ export function useMessages( getAttachment }; } - diff --git a/dashboard/src/i18n/locales/en-US/features/chat.json b/dashboard/src/i18n/locales/en-US/features/chat.json index 8f2505dec..26271d820 100644 --- a/dashboard/src/i18n/locales/en-US/features/chat.json +++ b/dashboard/src/i18n/locales/en-US/features/chat.json @@ -80,6 +80,14 @@ "today": "Today", "yesterday": "Yesterday" }, + "stats": { + "tokens": "Tokens", + "inputTokens": "Input Tokens", + "outputTokens": "Output Tokens", + "cachedTokens": "Cached Tokens", + "duration": "Duration", + "ttft": "Time to First Token" + }, "connection": { "title": "Connection Status Notice", "message": "The system detected that the chat connection needs to be re-established.", diff --git a/dashboard/src/i18n/locales/zh-CN/features/chat.json b/dashboard/src/i18n/locales/zh-CN/features/chat.json index 2ec8f51f9..2a51ef8bf 100644 --- a/dashboard/src/i18n/locales/zh-CN/features/chat.json +++ b/dashboard/src/i18n/locales/zh-CN/features/chat.json @@ -80,6 +80,14 @@ "today": "今天", "yesterday": "昨天" }, + "stats": { + "tokens": "Token", + "inputTokens": "输入 Token", + "outputTokens": "输出 Token", + "cachedTokens": "缓存 Token", + "duration": "耗时", + "ttft": "首字时间" + }, "connection": { "title": "连接状态提醒", "message": "系统检测到聊天连接需要重新建立。",