feat(http_llm): add ollama support

Author: Alchuang22-dev

config.json

@@ -18,7 +18,13 @@
             "is_detailed_tool_call": true,
             "is_detailed_observation": true
         },
-        "llm": {"temperature": 0.1, "max_tokens": 4096, "top_p": 1},
+        "llm": {    
+            "cls": "oxygent.llms.OllamaLLM",
+            "base_url": "http://localhost:11434",
+            "temperature": 0.1, 
+            "max_tokens": 4096, 
+            "top_p": 1
+        },
         "cache": {"save_dir": "./cache_dir"},
         "message": {
             "is_send_tool_call": true,

examples/agents/ollama_demo.py

@@ -0,0 +1,44 @@
+"""gemma_cli_demo.py
+"""
+
+import asyncio
+from oxygent import MAS, oxy  
+from oxygent.utils.env_utils import get_env_var
+
+oxy_space = [
+    oxy.HttpLLM(                
+        name="local_gemma",  
+        base_url="http://localhost:11434/api/chat", 
+        model_name=get_env_var("DEFAULT_OLLAMA_MODEL"),   
+        llm_params={"temperature": 0.2},    
+        semaphore=1,              
+        timeout=240,
+    ),
+
+    oxy.ChatAgent(
+        name="master_agent",
+        is_master=True,
+        llm_model="local_gemma",  
+    ),
+]
+
+async def chat():
+    async with MAS(oxy_space=oxy_space) as mas:
+        history = [{"role": "system", "content": "You are a helpful assistant."}]
+
+        while True:
+            user_in = input("User: ").strip()
+            if user_in.lower() in {"exit", "quit", "q"}:
+                break
+
+            history.append({"role": "user", "content": user_in})
+            result = await mas.call(
+                callee="master_agent",
+                arguments={"messages": history},
+            )
+            assistant_out = result
+            print(f"Assistant: {assistant_out}\n")
+            history.append({"role": "assistant", "content": assistant_out})
+
+if __name__ == "__main__":
+    asyncio.run(chat())

oxygent/config.py

@@ -53,7 +53,13 @@ class Config:
             "is_detailed_tool_call": True,
             "is_detailed_observation": True,
         },
-        "llm": {"temperature": 0.1, "max_tokens": 4096, "top_p": 1},
+        "llm": {    
+            "cls": "oxygent.llms.OllamaLLM",
+            "base_url": "http://localhost:11434",
+            "temperature": 0.1, 
+            "max_tokens": 4096, 
+            "top_p": 1
+        },
         "cache": {"save_dir": "./cache_dir"},
         "message": {
             "is_send_tool_call": True,

oxygent/oxy/llms/http_llm.py

@@ -13,6 +13,7 @@
 from ...schemas import OxyRequest, OxyResponse, OxyState
 from .remote_llm import RemoteLLM
 
+
 logger = logging.getLogger(__name__)
 
 
@@ -22,7 +23,7 @@ class HttpLLM(RemoteLLM):
     This class provides a concrete implementation of RemoteLLM for communicating
     with remote LLM APIs over HTTP. It handles API authentication, request
     formatting, and response parsing for OpenAI-compatible APIs.
-    """
+    """ 
 
     async def _execute(self, oxy_request: OxyRequest) -> OxyResponse:
         """Execute an HTTP request to the remote LLM API.
@@ -37,10 +38,11 @@ async def _execute(self, oxy_request: OxyRequest) -> OxyResponse:
         Returns:
             OxyResponse: The response containing the LLM's output with COMPLETED state.
         """
-        headers = {
-            "Content-Type": "application/json",
-            "Authorization": f"Bearer {self.api_key}",
-        }
+        use_openai = self.api_key is not None
+        url = self.base_url.rstrip("/")
+        headers = {"Content-Type": "application/json"}
+        if use_openai:
+            headers["Authorization"] = f"Bearer {self.api_key}"
 
         # Construct payload for the API request
         llm_config = Config.get_llm_config()
@@ -59,16 +61,21 @@ async def _execute(self, oxy_request: OxyRequest) -> OxyResponse:
 
         async with httpx.AsyncClient(timeout=self.timeout) as client:
             http_response = await client.post(
-                self.base_url, headers=headers, json=payload
+                url, headers=headers, json=payload
             )
             http_response.raise_for_status()
             data = http_response.json()
             if "error" in data:
                 error_message = data["error"].get("message", "Unknown error")
                 raise ValueError(f"LLM API error: {error_message}")
-            response_message = data["choices"][0]["message"]
-            result = response_message.get("content") or response_message.get(
-                "reasoning_content"
-            )
+            
+            if use_openai:
+                response_message = data["choices"][0]["message"]
+                result = response_message.get("content") or response_message.get(
+                    "reasoning_content"
+                )
+            else:  # ollama
+                result = data["message"]["content"]
+            
 
             return OxyResponse(state=OxyState.COMPLETED, output=result)

oxygent/oxy/llms/remote_llm.py

@@ -2,6 +2,7 @@
 
 from pydantic import Field, field_validator
 
+
 from ...schemas import OxyRequest, OxyResponse
 from .base_llm import BaseLLM
 
@@ -19,11 +20,11 @@ class RemoteLLM(BaseLLM):
         model_name: The specific model name to use for requests.
     """
 
-    api_key: Optional[str] = Field("")
+    api_key: Optional[str] = Field(default=None)
     base_url: Optional[str] = Field("")
     model_name: Optional[str] = Field("")
 
-    @field_validator("api_key", "base_url", "model_name")
+    @field_validator("base_url", "model_name")
     @classmethod
     def not_empty(cls, value, info):
         key = info.field_name
@@ -38,7 +39,7 @@ def not_empty(cls, value, info):
                 f"Environment variable '{key}' type error: expected str, got {type(value).__name__}."
             )
 
-        if not value.strip():
+        if not isinstance(value, str) or not value.strip():
             raise ValueError(f"{key} must be a non-empty string")
 
         return value