Welcome to Maev
Maev is autopilot and observability for AI agents in production. One line of code wrap gives you deep client-side intervention (circuit breakers, local tool interception) AND transparently routes traffic to our self-improving Gateway for automated prompt optimization.
import maev
# Wraps your agent natively, and routes LLM traffic to our Gateway
maev.run(my_agent, gateway=True)That is the entire integration.
What Maev does
Deep Client Control & Transparent Gateway. The moment you call maev.run(), Maev activates in-process protections (circuit breakers, loops, infinite execution guards). It simultaneously configures your agent's LLM client to route requests through gateway.maev.dev, letting the backend handle advanced self-reflection evaluations and fallback routing.
Observes everything automatically. Every LLM call, token count, cost, and latency is captured natively from the execution thread. Each run becomes a full trace timeline in the dashboard within seconds.
Gets smarter over time. After 20+ runs, Maev analyzes failure and success patterns, identifies strategies that work, and applies them automatically in future runs. Your agent improves without code changes.
Classifies failures in real time. Every session is scored against 10 failure categories including hallucinations, tool failures, infinite loops, cost anomalies, context exhaustion, and prompt injection. If something breaks, Maev classifies it, logs it, and alerts you.
Alerts via Slack and email. Get notified the moment a session fails, a cost threshold is crossed, or a critical failure is detected.
Tracks cost per agent. See what each agent costs per run and over time. Budget limits are enforced automatically at runtime via circuit breakers.