Switch LLM to Claude 3.5 Haiku via Anthropic

.env.example

@@ -1,4 +1,4 @@
 # Copy to .env locally and set Space secrets on Hugging Face (never commit .env)
-GOOGLE_API_KEY=your_google_api_key_here
-GEMINI_MODEL=gemini-2.0-flash
+ANTHROPIC_API_KEY=sk-ant-your_key_here
+ANTHROPIC_MODEL=claude-3-5-haiku-20241022
 QUESTION_DELAY_SECONDS=2

README.md

@@ -15,4 +15,4 @@ hf_oauth_expiration_minutes: 480
 
 LangGraph agent with web search for the Hugging Face agents course final assignment.
 
-**Setup:** Add `GOOGLE_API_KEY` in Space secrets. Use **Test agent** to try one question, then **Submit evaluation** after logging in with Hugging Face.
+**Setup:** Add `ANTHROPIC_API_KEY` in Space secrets (default model: `claude-3-5-haiku-20241022`). Log in with Hugging Face, then run evaluation submit.

agent.py

@@ -7,14 +7,34 @@ import re
 import json
 import time
 from dotenv import load_dotenv
-from langchain_google_genai import ChatGoogleGenerativeAI
-from langchain_google_genai.chat_models import ChatGoogleGenerativeAIError
+from langchain_anthropic import ChatAnthropic
 from typing_extensions import TypedDict
 from prompts import PLANNER_PROMPT, SYNTHESIS_ANSWER_PROMPT, ANSWER_PROMPT
 from langgraph.graph import StateGraph, START, END
 from tools import search_web
 
-load_dotenv()
+load_dotenv(override=True)
+
+
+def _resolve_anthropic_api_key() -> str:
+    """Use ANTHROPIC_API_KEY from .env (override=True), not IDE session tokens."""
+    key = (os.getenv("ANTHROPIC_API_KEY") or "").strip().strip('"').strip("'")
+    if not key:
+        raise ValueError(
+            "ANTHROPIC_API_KEY is not set. Add sk-ant-... to .env (see .env.example) "
+            "or Space secrets. Create keys at https://console.anthropic.com/settings/keys"
+        )
+    if key.startswith("sess-") or key.startswith("hf_"):
+        raise ValueError(
+            "ANTHROPIC_API_KEY must be an Anthropic key (sk-ant-...), not a Cursor/HF token."
+        )
+    if not key.startswith("sk-ant-"):
+        raise ValueError(
+            "ANTHROPIC_API_KEY should start with sk-ant-. "
+            "Create one at https://console.anthropic.com/settings/keys"
+        )
+    return key
+
 
 MAX_SEARCH_QUERIES = 3
 MIN_SEARCH_QUERIES = 1
@@ -24,9 +44,9 @@ QUESTION_DELAY_SECONDS = float(os.getenv("QUESTION_DELAY_SECONDS", "2"))
 
 #--- Model --------------------------------------------------------------#
 
-model = ChatGoogleGenerativeAI(
-    model=os.getenv("GEMINI_MODEL", "gemini-2.0-flash"),
-    api_key=os.getenv("GOOGLE_API_KEY"),
+model = ChatAnthropic(
+    model=os.getenv("ANTHROPIC_MODEL", "claude-3-5-haiku-20241022"),
+    api_key=_resolve_anthropic_api_key(),
     temperature=0,
 )
 
@@ -64,17 +84,40 @@ def _retry_delay_seconds(error: Exception, attempt: int) -> float:
 
 def _is_rate_limit_error(error: Exception) -> bool:
     msg = str(error).lower()
-    return "429" in msg or "resource_exhausted" in msg or "quota" in msg
+    return (
+        "429" in msg
+        or "rate_limit" in msg
+        or "rate limit" in msg
+        or "too many requests" in msg
+        or "resource_exhausted" in msg
+        or "quota" in msg
+        or "overloaded" in msg
+    )
+
+
+def _message_content(response) -> str:
+    content = response.content
+    if isinstance(content, str):
+        return content.strip()
+    if isinstance(content, list):
+        parts = []
+        for block in content:
+            if isinstance(block, str):
+                parts.append(block)
+            elif isinstance(block, dict) and block.get("text"):
+                parts.append(str(block["text"]))
+        return "\n".join(parts).strip()
+    return str(content).strip()
 
 
 def invoke_llm(prompt: str) -> str:
-    """Call Gemini with retries on rate-limit / quota errors."""
+    """Call Claude with retries on rate-limit errors."""
     last_error: Exception | None = None
     for attempt in range(MAX_LLM_RETRIES):
         try:
             response = model.invoke(prompt)
-            return response.content.strip()
-        except ChatGoogleGenerativeAIError as e:
+            return _message_content(response)
+        except Exception as e:
             last_error = e
             if _is_rate_limit_error(e) and attempt < MAX_LLM_RETRIES - 1:
                 delay = _retry_delay_seconds(e, attempt)
@@ -217,7 +260,7 @@ def plan_node(state: AgentState) -> AgentState:
     try:
         content = invoke_llm(prompt)
         data = _parse_planner_response(content, question)
-    except (json.JSONDecodeError, TypeError, KeyError, ChatGoogleGenerativeAIError):
+    except Exception:
         data = {
             "steps": [
                 "Identify entities and constraints in the question",
@@ -255,12 +298,12 @@ def finalize_node(state: AgentState) -> AgentState:
         )
         parsed = _parse_synthesis_answer(invoke_llm(prompt))
         return parsed
-    except (json.JSONDecodeError, TypeError, KeyError, ValueError, ChatGoogleGenerativeAIError) as e:
+    except Exception as e:
         print(f"Finalize LLM failed ({e}); using heuristic synthesis.")
         notes = _heuristic_synthesis(search_results)
         try:
             answer = _answer_from_notes(question, plan, notes)
-        except ChatGoogleGenerativeAIError:
+        except Exception:
             first_line = notes.split("\n", 1)[0].lstrip("- ").strip()
             answer = first_line[:200] if first_line else "Unknown"
         return {"synthesis_notes": notes, "answer": answer}
@@ -286,7 +329,7 @@ def run_agent(question: str) -> AgentState:
 
 
 if __name__ == "__main__":
-    # Free tier: ~20 requests/day per model — use 1 question while testing.
+    # Use 1 question while testing to save API usage.
     questions = [
         "What is the capital of France?",
     ]

app.py

@@ -2,7 +2,7 @@ import os
 
 from dotenv import load_dotenv
 
-load_dotenv()
+load_dotenv(override=True)
 
 import gradio as gr
 import requests
@@ -185,12 +185,12 @@ if __name__ == "__main__":
     # Check for SPACE_HOST and SPACE_ID at startup for information
     space_host_startup = os.getenv("SPACE_HOST")
     space_id_startup = os.getenv("SPACE_ID") # Get SPACE_ID at startup
-    GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY")
-    
-    if GOOGLE_API_KEY:
-        print(f"✅ GOOGLE_API_KEY found: {GOOGLE_API_KEY}")
+    anthropic_api_key = os.getenv("ANTHROPIC_API_KEY")
+
+    if anthropic_api_key:
+        print("✅ ANTHROPIC_API_KEY found (set).")
     else:
-        print("ℹ️  GOOGLE_API_KEY environment variable not found (running locally?).")
+        print("ℹ️  ANTHROPIC_API_KEY environment variable not found (running locally?).")
 
     if space_host_startup:
         print(f"✅ SPACE_HOST found: {space_host_startup}")

requirements.txt

@@ -3,9 +3,7 @@ python-dotenv
 requests
 langgraph
 langchain
-langchain-google-genai
+langchain-anthropic
 ddgs
-google-generativeai
-google-auth
 gradio[oauth]
 ipython