Add openai_compatible provider branch for freellmapi gateway

ChatOpenAI(base_url, api_key) path with no client-side rate limiter (gateway handles failover + limits). Set LLM_PROVIDER=openai_compatible + the OPENAI_COMPATIBLE_* vars to use it.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

.env.example

@@ -4,15 +4,20 @@ TAVILY_API_KEY=
 # Required when LLM_PROVIDER=gemini (default). Free key: aistudio.google.com
 GOOGLE_API_KEY=
 
-# --- LLM provider for text/reasoning + judge nodes: "gemini" or "groq" ---
-# Default "gemini" + gemini-2.5-flash (best quality, high free-tier TPM).
-# GOOGLE_API_KEY must come from a project WITHOUT billing (else free tier = 0).
+# --- LLM provider for text/reasoning + judge nodes ---
+# "openai_compatible" (freellmapi gateway, best for rate limits), "gemini", "groq".
 LLM_PROVIDER=gemini
 GEMINI_TEXT_MODEL=gemini-2.5-flash
 GROQ_TEXT_MODEL=llama-3.1-8b-instant
 # Global pace across all text-LLM calls (req/sec). 0.15 = ~9/min (<10 RPM free).
+# Ignored when LLM_PROVIDER=openai_compatible (gateway manages its own limits).
 RATE_LIMIT_RPS=0.15
 
+# --- freellmapi gateway (set LLM_PROVIDER=openai_compatible to use) ---
+OPENAI_COMPATIBLE_BASE_URL=
+OPENAI_COMPATIBLE_API_KEY=
+OPENAI_COMPATIBLE_MODEL=auto
+
 # --- Optional: LangSmith tracing (nice for debugging the graph) ---
 # Set TRACING=true and provide the API key to stream runs to smith.langchain.com.
 # LangChain/LangGraph pick these up automatically; no code change needed.

gaia_agent/config.py

@@ -24,13 +24,18 @@ class Settings(BaseSettings):
     google_api_key: str = ""
 
     # --- LLM provider for the text/reasoning + judge nodes ---
-    # "gemini" (default, best quality + high free-tier TPM) or "groq".
+    # "openai_compatible" (freellmapi gateway, recommended), "gemini", or "groq".
     # Vision + Whisper always use Groq.
     llm_provider: str = "gemini"
 
     # --- Models (override via env) ---
     gemini_text_model: str = "gemini-2.5-flash"
     groq_text_model: str = "llama-3.1-8b-instant"
+
+    # --- OpenAI-compatible gateway (freellmapi) ---
+    openai_compatible_base_url: str = ""  # e.g. https://my-gateway.example.com/v1
+    openai_compatible_api_key: str = ""
+    openai_compatible_model: str = "auto"
     groq_vision_model: str = "meta-llama/llama-4-scout-17b-16e-instruct"
     groq_whisper_model: str = "whisper-large-v3"
 

gaia_agent/llm.py

@@ -31,7 +31,22 @@ def get_text_llm(temperature: float = 0.0):
     s = get_settings()
     limiter = _rate_limiter()
 
-    if s.llm_provider.lower() == "groq":
+    provider = s.llm_provider.lower()
+
+    if provider == "openai_compatible":
+        # freellmapi gateway: it does its own multi-provider failover + rate
+        # tracking, so no client-side rate limiter is applied here.
+        from langchain_openai import ChatOpenAI
+
+        return ChatOpenAI(
+            model=s.openai_compatible_model,
+            base_url=s.openai_compatible_base_url,
+            api_key=s.openai_compatible_api_key,
+            temperature=temperature,
+            max_retries=3,
+        )
+
+    if provider == "groq":
         from langchain_groq import ChatGroq
 
         return ChatGroq(

pyproject.toml

@@ -15,6 +15,7 @@ dependencies = [
     "langchain-core>=0.3.0",
     "langchain-groq>=0.2.0",
     "langchain-google-genai>=2.0.0",
+    "langchain-openai>=0.2.0",
     "langchain-tavily>=0.1.0",
     "langchain-community>=0.3.0",
     "wikipedia>=1.4.0",

requirements.txt

@@ -9,6 +9,7 @@ langchain>=0.3.0
 langchain-core>=0.3.0
 langchain-groq>=0.2.0
 langchain-google-genai>=2.0.0
+langchain-openai>=0.2.0
 langchain-tavily>=0.1.0
 langchain-community>=0.3.0
 wikipedia