Diagnose and fix your AI agent.
Without the code.

# Add these two lines, that's it
import agent_doctor
agent_doctor.init(api_key="sk-proj-...")

# Your existing agent code, zero changes needed
# LLM calls, vector DB queries and tool calls are auto-captured

def handle_ticket(query, customer_id):
    # LLM call → auto-captured
    intent = openai.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": query}]
    )

    # Vector DB lookup → auto-captured
    docs = pinecone_index.query(vector=embed(query), top_k=5)

    # Tool call → auto-captured
    policy = check_refund_policy(customer_id)

    # LLM call → auto-captured
    response = openai.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "system", "content": f"Docs: {docs} Policy: {policy}"},
                  {"role": "user", "content": query}]
    )
    return response

# Webhook handler. Overrides auto-injected by the SDK
with agent_doctor.run_context(run_id=payload.run_id, overrides=payload.overrides):
    result = handle_ticket(payload.query, payload.customer_id)
    # Every step auto-captured. Overrides auto-injected. Done.