AstrBot 4.10.2__py3-none-any.whl → 4.10.4__py3-none-any.whl

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/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}"
+            )