feat: support Gemma 4 chat and agent trials

SovereignAI-Starter/app/main.py

@@ -47,6 +47,10 @@ class ChatRequest(BaseModel):
 
 class AgentRequest(BaseModel):
     task: str = Field(description="مهمة قصيرة للوكيل المحلي")
+    model: str | None = Field(
+        default=None,
+        description="اسم نموذج Ollama؛ اتركه فارغًا لاستخدام النموذج الافتراضي",
+    )
 
 
 class StoredMessage(BaseModel):
@@ -106,7 +110,7 @@ def get_local_user() -> dict[str, str]:
 
 def chat_payload(request: ChatRequest, *, stream: bool) -> dict[str, Any]:
     model = request.model or os.getenv("LOCAL_MODEL", "qwen2.5:1.5b-instruct-q4_K_M")
-    return {
+    payload: dict[str, Any] = {
         "model": model,
         "messages": (
             [message.model_dump() for message in request.messages]
@@ -130,6 +134,10 @@ def chat_payload(request: ChatRequest, *, stream: bool) -> dict[str, Any]:
         ),
         "stream": stream,
     }
+    # Disable Gemma 4 thinking so Ollama places the answer in `content` for clients.
+    if model.lower().startswith("gemma4"):
+        payload["reasoning_effort"] = "none"
+    return payload
 
 
 @app.post("/v1/chat/completions")
@@ -283,7 +291,7 @@ async def run_agent(request: AgentRequest) -> dict[str, Any]:
         pass
 
     base_url = os.getenv("LOCAL_LLM_BASE_URL", "http://127.0.0.1:11434/v1").rstrip("/")
-    model = os.getenv("LOCAL_MODEL", "qwen2.5:1.5b-instruct-q4_K_M")
+    model = request.model or os.getenv("LOCAL_MODEL", "qwen2.5:1.5b-instruct-q4_K_M")
     payload = {
         "model": model,
         "messages": [
@@ -292,6 +300,8 @@ async def run_agent(request: AgentRequest) -> dict[str, Any]:
         ],
         "stream": False,
     }
+    if model.lower().startswith("gemma4"):
+        payload["reasoning_effort"] = "none"
     completion = await get_completion(payload, base_url)
     return {
         "task": request.task,