fix: route LLM to Groq (deepseek-r1) instead of hardcoded gpt-4o/openai

agent/reflection_agent.py

@@ -455,28 +455,50 @@ def _call_llm(
     client=None,
     model: str = "gpt-4o",
 ) -> tuple[str, dict]:
-    """Call OpenAI chat completion. Returns (patch_text, usage_dict)."""
+    """
+    Call the configured LLM provider (Groq, OpenAI, etc.).
+    Auto-detects provider from settings when client is None.
+    Returns (patch_text, usage_dict).
+    """
+    from configs.settings import settings
+
+    provider = settings.llm_provider.lower()
+    effective_model = model
+
+    # ── Groq (free, recommended) ───────────────────────────────────────────
+    if client is None and provider == "groq":
+        try:
+            from groq import Groq
+            client = Groq(api_key=settings.groq_api_key)
+            effective_model = settings.llm_model  # use configured Groq model
+        except ImportError as e:
+            raise ImportError("Install groq: pip install groq") from e
+
+    # ── OpenAI (fallback) ─────────────────────────────────────────────────
     if client is None:
         try:
             from openai import OpenAI
-            client = OpenAI()
+            client = OpenAI(api_key=settings.openai_api_key or None)
         except ImportError as e:
-            raise ImportError("Install openai: pip install openai") from e
+            raise ImportError(
+                "No LLM client available. Set LLM_PROVIDER=groq and GROQ_API_KEY, "
+                "or install openai: pip install openai"
+            ) from e
 
     response = client.chat.completions.create(
-        model=model,
+        model=effective_model,
         messages=[
             {"role": "system", "content": SYSTEM_PROMPT},
-            {"role": "user", "content": user_prompt},
+            {"role": "user",   "content": user_prompt},
         ],
-        max_tokens=4096,
-        temperature=0.2,
+        max_tokens=settings.llm_max_tokens,
+        temperature=settings.llm_temperature,
     )
     patch_text = response.choices[0].message.content or ""
     usage = {
-        "prompt_tokens": response.usage.prompt_tokens,
+        "prompt_tokens":     response.usage.prompt_tokens,
         "completion_tokens": response.usage.completion_tokens,
-        "total_tokens": response.usage.total_tokens,
+        "total_tokens":      response.usage.total_tokens,
     }
     return patch_text, usage
 

api/tasks.py

@@ -165,9 +165,10 @@ async def run_agent_task_async(
         traj_path = Path(f"results/trajectories/{task_id}.jsonl")
         traj_logger = TrajectoryLogger(traj_path)
 
+        from configs.settings import settings
         from agent.reflection_agent import ReflectionAgent
         agent = ReflectionAgent(
-            model="gpt-4o",
+            model=settings.llm_model,   # reads LLM_MODEL from env (e.g. deepseek-r1-distill-llama-70b)
             max_attempts=max_attempts,
             sandbox=sandbox,
             trajectory_logger=traj_logger,

configs/settings.py

@@ -18,6 +18,8 @@ class Settings(BaseSettings):
 
     # ── LLM ─────────────────────────────────────────────────────────────────
     openai_api_key: str = Field(default="", alias="OPENAI_API_KEY")
+    groq_api_key: str = Field(default="", alias="GROQ_API_KEY")
+    llm_provider: str = Field(default="openai", alias="LLM_PROVIDER")  # openai | groq | gemini | ollama
     llm_model: str = Field(default="gpt-4o", alias="LLM_MODEL")
     llm_max_tokens: int = Field(default=4096, alias="LLM_MAX_TOKENS")
     llm_temperature: float = Field(default=0.2, alias="LLM_TEMPERATURE")