I started Beacon because while building AI agents, I kept running into the same problem: I could see the final output, but understanding how the agent got there meant going through logs and traces afterwards. I wanted to see the execution itself. That led me to building Beacon around OpenTelemetry — taking the traces an agent already produces and turning them into a real-time execution DAG. Once I had that working, replay felt like a natural next step, so you could go back through a completed run and understand exactly how it unfolded. The project also became a fun systems problem along the way — OTLP ingestion, async processing with BullMQ, PostgreSQL for run history, Redis Pub/Sub and WebSockets for realtime updates, and React Flow for the execution graph. It started as something I wanted for my own agents, and I’m launching it now to see how other builders use it and what I should build next.

Beacon

See your AI agents execute in real time.

See your AI agents execute in real time.