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

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 +359,27 @@ 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"
+                message["reasoning_content"] = reasoning_content
+
     async def _handle_api_error(
         self,
         e: Exception,
@@ -476,6 +473,7 @@ 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 +483,7 @@ class ProviderOpenAIOfficial(Provider):
             system_prompt,
             tool_calls_result,
             model=model,
+            extra_user_content_parts=extra_user_content_parts,
             **kwargs,
         )
 
@@ -624,33 +623,71 @@ 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/context.py

@@ -377,7 +377,7 @@ 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):