feat: add Ollama and Claude adapters to LLM preset adapters

Author: lss233

plugins/llm_preset_adapters/__init__.py

@@ -4,16 +4,21 @@
 from llm_preset_adapters.gemini_adapter import GeminiAdapter, GeminiConfig
 from llm_preset_adapters.deepseek_adapter import DeepSeekAdapter, DeepSeekConfig
 from llm_preset_adapters.openai_adapter import OpenAIAdapter, OpenAIConfig
+from llm_preset_adapters.ollama_adapter import OllamaAdapter, OllamaConfig
+from llm_preset_adapters.claude_adapter import ClaudeAdapter, ClaudeConfig
 
 logger = get_logger("LLMPresetAdapters")
 class LLMPresetAdaptersPlugin(Plugin):
     def __init__(self):
         pass
 
+
     def on_load(self):
         self.llm_registry.register("openai", OpenAIAdapter, OpenAIConfig, LLMAbility.TextChat)
         self.llm_registry.register("deepseek", DeepSeekAdapter, DeepSeekConfig, LLMAbility.TextChat)
         self.llm_registry.register("gemini", GeminiAdapter, GeminiConfig, LLMAbility.TextChat)
+        self.llm_registry.register("ollama", OllamaAdapter, OllamaConfig, LLMAbility.TextChat)
+        self.llm_registry.register("claude", ClaudeAdapter, ClaudeConfig, LLMAbility.TextChat)
         logger.info("LLMPresetAdaptersPlugin loaded")
 
     def on_start(self):

plugins/llm_preset_adapters/claude_adapter.py

@@ -0,0 +1,96 @@
+from pydantic import ConfigDict, BaseModel
+import requests
+from framework.llm.adapter import LLMBackendAdapter
+from framework.llm.format.request import LLMChatRequest
+from framework.llm.format.response import LLMChatResponse
+from framework.logger import get_logger
+
+class ClaudeConfig(BaseModel):
+    api_key: str
+    api_base: str = "https://api.anthropic.com/v1"
+    model_config = ConfigDict(frozen=True)
+
+def convert_messages_to_claude_prompt(messages) -> str:
+    """将消息列表转换为 Claude 的对话格式"""
+    prompt = ""
+    for msg in messages:
+        if msg.role == "system":
+            # Claude 没有专门的系统消息，我们将其作为 Human 的第一条消息
+            prompt += f"Human: {msg.content}\n\nAssistant: I understand. I'll follow these instructions.\n\n"
+        elif msg.role == "user":
+            prompt += f"Human: {msg.content}\n\n"
+        elif msg.role == "assistant":
+            prompt += f"Assistant: {msg.content}\n\n"
+    # 添加最后的 Assistant: 前缀来获取回复
+    prompt += "Assistant: "
+    return prompt
+
+class ClaudeAdapter(LLMBackendAdapter):
+    def __init__(self, config: ClaudeConfig):
+        self.config = config
+        self.logger = get_logger("ClaudeAdapter")
+
+    def chat(self, req: LLMChatRequest) -> LLMChatResponse:
+        api_url = f"{self.config.api_base}/messages"
+        headers = {
+            "x-api-key": self.config.api_key,
+            "anthropic-version": "2023-06-01",
+            "content-type": "application/json"
+        }
+
+        # 构建请求数据
+        data = {
+            "model": req.model,
+            "messages": [
+                {
+                    "role": "user" if msg.role == "user" else "assistant",
+                    "content": msg.content
+                }
+                for msg in req.messages
+                if msg.role in ["user", "assistant"]  # 跳过 system 消息，因为 Claude API 不支持
+            ],
+            "max_tokens": req.max_tokens,
+            "temperature": req.temperature,
+            "top_p": req.top_p,
+            "stream": req.stream
+        }
+
+        # 如果有系统消息，将其添加到第一个用户消息前面
+        system_messages = [msg for msg in req.messages if msg.role == "system"]
+        if system_messages:
+            if len(data["messages"]) > 0 and data["messages"][0]["role"] == "user":
+                data["messages"][0]["content"] = f"{system_messages[0].content}\n\n{data['messages'][0]['content']}"
+
+        # Remove None fields
+        data = {k: v for k, v in data.items() if v is not None}
+        
+        response = requests.post(api_url, json=data, headers=headers)
+        try:
+            response.raise_for_status()
+            response_data = response.json()
+        except Exception as e:
+            self.logger.error(f"API Response: {response.text}")
+            raise e
+
+        # 转换 Claude 响应格式为标准的 LLMChatResponse 格式
+        transformed_response = {
+            "id": response_data.get("id", ""),
+            "object": "chat.completion",
+            "created": response_data.get("created_at", 0),
+            "model": req.model,
+            "choices": [{
+                "index": 0,
+                "message": {
+                    "role": "assistant",
+                    "content": response_data["content"][0]["text"]
+                },
+                "finish_reason": response_data.get("stop_reason", "stop")
+            }],
+            "usage": {
+                "prompt_tokens": 0,  # Claude API 目前不返回 token 使用量
+                "completion_tokens": 0,
+                "total_tokens": 0
+            }
+        }
+        
+        return LLMChatResponse(**transformed_response) 

plugins/llm_preset_adapters/ollama_adapter.py

@@ -0,0 +1,77 @@
+from pydantic import ConfigDict, BaseModel
+import requests
+from framework.llm.adapter import LLMBackendAdapter
+from framework.llm.format.request import LLMChatRequest
+from framework.llm.format.response import LLMChatResponse
+from framework.logger import get_logger
+
+class OllamaConfig(BaseModel):
+    api_base: str = "http://localhost:11434"
+    model_config = ConfigDict(frozen=True)
+
+class OllamaAdapter(LLMBackendAdapter):
+    def __init__(self, config: OllamaConfig):
+        self.config = config
+        self.logger = get_logger("OllamaAdapter")
+
+    def chat(self, req: LLMChatRequest) -> LLMChatResponse:
+        api_url = f"{self.config.api_base}/api/chat"
+        headers = {
+            "Content-Type": "application/json"
+        }
+
+        # 将消息转换为 Ollama 格式
+        messages = []
+        for msg in req.messages:
+            messages.append({
+                "role": msg.role,
+                "content": msg.content
+            })
+
+        data = {
+            "model": req.model,
+            "messages": messages,
+            "stream": req.stream,
+            "options": {
+                "temperature": req.temperature,
+                "top_p": req.top_p,
+                "num_predict": req.max_tokens,
+                "stop": req.stop
+            }
+        }
+
+        # Remove None fields
+        data = {k: v for k, v in data.items() if v is not None}
+        if "options" in data:
+            data["options"] = {k: v for k, v in data["options"].items() if v is not None}
+        
+        response = requests.post(api_url, json=data, headers=headers)
+        try:
+            response.raise_for_status()
+            response_data = response.json()
+        except Exception as e:
+            print(f"API Response: {response.text}")
+            raise e
+        
+        # 转换 Ollama 响应格式为标准的 LLMChatResponse 格式
+        transformed_response = {
+            "id": "ollama-" + req.model,
+            "object": "chat.completion",
+            "created": 0,
+            "model": req.model,
+            "choices": [{
+                "index": 0,
+                "message": {
+                    "role": "assistant",
+                    "content": response_data["message"]["content"]
+                },
+                "finish_reason": "stop"
+            }],
+            "usage": {
+                "prompt_tokens": 0,
+                "completion_tokens": 0,
+                "total_tokens": 0
+            }
+        }
+        
+        return LLMChatResponse(**transformed_response) 

plugins/llm_preset_adapters/setup.py

@@ -9,7 +9,8 @@
     install_requires=[
         "openai",
         "google-generativeai",
-        "anthropic"
+        "anthropic",
+        "requests"
     ],
     entry_points={
         "chatgpt_mirai.plugins": [