Switch to HuggingFace API (Groq rate limited)

agent.py

@@ -25,7 +25,7 @@ from langchain_community.tools import DuckDuckGoSearchRun
 from langchain_community.document_loaders import WikipediaLoader
 from langchain_community.document_loaders import ArxivLoader
 from langgraph.prebuilt import ToolNode, tools_condition
-from langchain_huggingface import HuggingFaceEmbeddings
+from langchain_huggingface import HuggingFaceEmbeddings, ChatHuggingFace, HuggingFaceEndpoint
 from langchain_groq import ChatGroq
 from langchain_community.vectorstores import SupabaseVectorStore
 from langchain_core.messages import SystemMessage, HumanMessage
@@ -741,12 +741,21 @@ tools = [
 def build_graph():
     """Build the graph"""
     
-    # Use Groq (fast, reliable, free tier)
-    # Model: qwen/qwen3-32b works well with LangChain tool calling
-    llm = ChatGroq(
-        model="qwen/qwen3-32b",
-        temperature=0,
-        api_key=os.environ.get("GROQ_API_KEY")
+    # Try Groq first, fall back to HuggingFace if rate limited
+    groq_key = os.environ.get("GROQ_API_KEY")
+    hf_token = os.environ.get("HF_TOKEN")
+    
+    # Use HuggingFace as primary (more reliable, no strict daily limits)
+    llm = ChatHuggingFace(
+        llm=HuggingFaceEndpoint(
+            repo_id="Qwen/Qwen2.5-72B-Instruct",
+            task="text-generation",
+            max_new_tokens=4096,
+            do_sample=False,
+            temperature=0.01,
+            huggingfacehub_api_token=hf_token
+        ),
+        verbose=False,
     )
 
     # Bind tools to LLM