LangGraph REPL + WebLLM Demo

Model: Ready to load model

# LangGraph-style Agent Loop Example
# Demonstrates an agent that iteratively refines its understanding
from js import ask_model, should_stop

class AgentState:
    def __init__(self, topic):
        self.topic = topic
        self.iteration = 0
        self.max_iterations = 3
        self.knowledge = []
        self.next_question = f"What is {topic}?"

def should_continue(self):
        return self.iteration < self.max_iterations and not should_stop()

# Agent nodes
async def research_node(state):
    """Research node: ask a question and gather knowledge"""
    print(f"\n{'='*60}")
    print(f"🔍 Iteration {state.iteration + 1}/{state.max_iterations}")
    print(f"Question: {state.next_question}")
    print(f"{'='*60}")

# Call the LLM (will throw error if stopped)
    answer = await ask_model(state.next_question)
    print(f"Answer: {answer}\n")

# Update state
    state.knowledge.append({
        "question": state.next_question,
        "answer": answer
    })
    state.iteration += 1

return state

async def planning_node(state):
    """Planning node: decide what to ask next"""
    if state.should_continue():
        # Generate next question based on current knowledge
        prompt = f"Based on this answer about {state.topic}: '{state.knowledge[-1]['answer']}', ask ONE brief follow-up question to learn more. Only output the question, nothing else."
        state.next_question = await ask_model(prompt)
        print(f"💭 Next question planned: {state.next_question}\n")

return state

async def summary_node(state):
    """Summary node: synthesize all knowledge"""
    print(f"\n{'='*60}")
    print(f"📚 FINAL SUMMARY - {state.topic}")
    print(f"{'='*60}")
    print(f"Gathered {len(state.knowledge)} pieces of knowledge:\n")

for i, item in enumerate(state.knowledge, 1):
        print(f"{i}. Q: {item['question']}")
        print(f"   A: {item['answer'][:100]}...\n")

return state

# Main agent loop (LangGraph-style)
async def run_agent(topic):
    state = AgentState(topic)

print(f"🤖 Starting LangGraph Agent for topic: '{topic}'")

# Agent loop
    while state.should_continue():
        state = await research_node(state)
        state = await planning_node(state)

# Final summary
    state = await summary_node(state)

return state

# Run the agent
result = await run_agent("Python programming")
print(f"\n✅ Agent completed {result.iteration} iterations")