Feat/add litellm provider

docutranslate/agents/agent.py

@@ -439,6 +439,44 @@ def _add_thinking_mode(self, data: dict):
             # 普通字段直接设置
             data[field_thinking] = value
 
+    def _call_litellm_sync(self, data: dict) -> dict:
+        """Call litellm.completion() synchronously. Returns OpenAI-format response dict."""
+        import litellm
+        kwargs = {
+            "model": data["model"],
+            "messages": data["messages"],
+            "temperature": data.get("temperature"),
+            "top_p": data.get("top_p"),
+            "stream": False,
+            "drop_params": True,
+        }
+        if self.key and self.key != "xx":
+            kwargs["api_key"] = self.key
+        if data.get("response_format"):
+            kwargs["response_format"] = data["response_format"]
+        kwargs = {k: v for k, v in kwargs.items() if v is not None}
+        response = litellm.completion(**kwargs)
+        return response.model_dump()
+
+    async def _call_litellm_async(self, data: dict) -> dict:
+        """Call litellm.acompletion() asynchronously. Returns OpenAI-format response dict."""
+        import litellm
+        kwargs = {
+            "model": data["model"],
+            "messages": data["messages"],
+            "temperature": data.get("temperature"),
+            "top_p": data.get("top_p"),
+            "stream": False,
+            "drop_params": True,
+        }
+        if self.key and self.key != "xx":
+            kwargs["api_key"] = self.key
+        if data.get("response_format"):
+            kwargs["response_format"] = data["response_format"]
+        kwargs = {k: v for k, v in kwargs.items() if v is not None}
+        response = await litellm.acompletion(**kwargs)
+        return response.model_dump()
+
     def _prepare_request_data(
             self, prompt: str, system_prompt: str, temperature=None, top_p=None, json_format=False
     ):
@@ -526,14 +564,17 @@ async def _continue_fetch_async(
         headers, data = self._prepare_request_data(continue_prompt, system_prompt, json_format=force_json)
 
         try:
-            response = await client.post(
-                f"{self.baseurl}/chat/completions",
-                json=data,
-                headers=headers,
-                timeout=self.timeout,
-            )
-            response.raise_for_status()
-            response_data = _parse_response_json(response)
+            if self.provider == "litellm":
+                response_data = await self._call_litellm_async(data)
+            else:
+                response = await client.post(
+                    f"{self.baseurl}/chat/completions",
+                    json=data,
+                    headers=headers,
+                    timeout=self.timeout,
+                )
+                response.raise_for_status()
+                response_data = _parse_response_json(response)
 
             # 安全提取 choices 和 content
             choices = response_data.get("choices", [])
@@ -640,14 +681,17 @@ async def send_async(
         input_tokens = 0
         output_tokens = 0
         try:
-            response = await client.post(
-                f"{self.baseurl}/chat/completions",
-                json=data,
-                headers=headers,
-                timeout=self.timeout,
-            )
-            response.raise_for_status()
-            response_data = _parse_response_json(response)
+            if self.provider == "litellm":
+                response_data = await self._call_litellm_async(data)
+            else:
+                response = await client.post(
+                    f"{self.baseurl}/chat/completions",
+                    json=data,
+                    headers=headers,
+                    timeout=self.timeout,
+                )
+                response.raise_for_status()
+                response_data = _parse_response_json(response)
 
             # 检查 finish_reason
             choices = response_data.get("choices", [])
@@ -950,14 +994,17 @@ def _continue_fetch(
         headers, data = self._prepare_request_data(continue_prompt, system_prompt, json_format=force_json)
 
         try:
-            response = client.post(
-                f"{self.baseurl}/chat/completions",
-                json=data,
-                headers=headers,
-                timeout=self.timeout,
-            )
-            response.raise_for_status()
-            response_data = _parse_response_json(response)
+            if self.provider == "litellm":
+                response_data = self._call_litellm_sync(data)
+            else:
+                response = client.post(
+                    f"{self.baseurl}/chat/completions",
+                    json=data,
+                    headers=headers,
+                    timeout=self.timeout,
+                )
+                response.raise_for_status()
+                response_data = _parse_response_json(response)
 
             # 安全提取 choices 和 content
             choices = response_data.get("choices", [])
@@ -1058,14 +1105,17 @@ def send(
         current_partial_result = None
 
         try:
-            response = client.post(
-                f"{self.baseurl}/chat/completions",
-                json=data,
-                headers=headers,
-                timeout=self.timeout,
-            )
-            response.raise_for_status()
-            response_data = _parse_response_json(response)
+            if self.provider == "litellm":
+                response_data = self._call_litellm_sync(data)
+            else:
+                response = client.post(
+                    f"{self.baseurl}/chat/completions",
+                    json=data,
+                    headers=headers,
+                    timeout=self.timeout,
+                )
+                response.raise_for_status()
+                response_data = _parse_response_json(response)
 
             # 检查 finish_reason
             choices = response_data.get("choices", [])

docutranslate/agents/provider/provider.py

@@ -1,6 +1,6 @@
 from typing import TypeAlias, Literal
 
-ProviderType: TypeAlias = Literal["minimax", "ollama", "bigmodel", "aliyuncs", "volces", "google", "siliconflow", "deepseek", "default"]
+ProviderType: TypeAlias = Literal["minimax", "ollama", "bigmodel", "aliyuncs", "volces", "google", "siliconflow", "deepseek", "litellm", "default"]
 
 def get_provider_by_domain(domain:str)->ProviderType:
     if domain == "open.bigmodel.cn":

docutranslate/agents/thinking/thinking_factory.py

@@ -2,7 +2,7 @@
 
 from docutranslate.agents.provider import ProviderType
 
-ModeType: TypeAlias = Literal["ollama", "bigmodel", "aliyuncs", "volces", "google", "siliconflow", "deepseek", "default"]
+ModeType: TypeAlias = Literal["ollama", "bigmodel", "aliyuncs", "volces", "google", "siliconflow", "deepseek", "litellm", "default"]
 ThinkingField: TypeAlias = str
 EnableValueType: TypeAlias = str | dict[str, Any] | bool
 DisableValueType: TypeAlias = str | dict[str, Any] | bool
@@ -23,6 +23,7 @@
     "google": ("reasoning_effort", "medium", "none"),
     "siliconflow": ("enable_thinking", True, False),
     "deepseek": ("thinking", {"type": "enabled"}, {"type": "disabled"}),
+    "litellm": ("reasoning_effort", "medium", "none"),
     "default": ("reasoning_effort", "medium", "none"),
 }
 
@@ -55,4 +56,6 @@ def get_thinking_mode(provider: ProviderType, model_id: str) -> ThinkingConfig:
         return thinking_mode["ollama"]
     elif provider == "deepseek":
         return thinking_mode["deepseek"]
+    elif provider == "litellm":
+        return get_thinking_mode_by_model_id(model_id)
     return get_thinking_mode_by_model_id(model_id)

tests/test_litellm_provider.py

@@ -0,0 +1,50 @@
+"""Tests for LiteLLM provider integration."""
+
+from docutranslate.agents.provider.provider import get_provider_by_domain, ProviderType
+from docutranslate.agents.thinking.thinking_factory import get_thinking_mode, thinking_mode
+
+
+class TestProviderDetection:
+
+    def test_litellm_not_autodetected(self):
+        """LiteLLM runs on any host - it's set via --provider litellm, not auto-detected."""
+        assert get_provider_by_domain("localhost") == "default"
+        assert get_provider_by_domain("127.0.0.1") == "default"
+        assert get_provider_by_domain("my-proxy.internal") == "default"
+
+    def test_existing_providers_unaffected(self):
+        assert get_provider_by_domain("open.bigmodel.cn") == "bigmodel"
+        assert get_provider_by_domain("dashscope.aliyuncs.com") == "aliyuncs"
+        assert get_provider_by_domain("api.deepseek.com") == "deepseek"
+        assert get_provider_by_domain("generativelanguage.googleapis.com") == "google"
+        assert get_provider_by_domain("api.siliconflow.cn") == "siliconflow"
+        assert get_provider_by_domain("ark.cn-beijing.volces.com") == "volces"
+
+
+class TestThinkingMode:
+
+    def test_litellm_in_thinking_mode_dict(self):
+        assert "litellm" in thinking_mode
+
+    def test_litellm_thinking_delegates_to_model_id(self):
+        """LiteLLM proxies multiple providers, so thinking mode should be
+        resolved by model name rather than a fixed provider config."""
+        result = get_thinking_mode("litellm", "qwen-plus")
+        assert result is not None
+        assert result == thinking_mode["aliyuncs"]
+
+    def test_litellm_thinking_gemini_model(self):
+        result = get_thinking_mode("litellm", "gemini-2.5-flash")
+        assert result == thinking_mode["google"]
+
+    def test_litellm_thinking_glm_model(self):
+        result = get_thinking_mode("litellm", "glm-4-plus")
+        assert result == thinking_mode["bigmodel"]
+
+    def test_litellm_thinking_unknown_model_uses_default(self):
+        result = get_thinking_mode("litellm", "some-random-model")
+        assert result == thinking_mode["default"]
+
+    def test_litellm_type_in_provider_type(self):
+        provider: ProviderType = "litellm"
+        assert provider == "litellm"