Hey Product Hunt 👋 — Leo here, co-founder of Agent Cockpit.

The honest origin story: I had four terminal tabs open on a Tuesday. One Claude Code session waiting for tool approval that I didn't notice for 20 minutes. One Codex run stuck in a loop, burning tokens. One that had finished an hour earlier and was just sitting there idle. One actually doing work.

I only found each of them when I happened to switch to the right tab.

That's the problem we built Agent Cockpit to solve. The more you lean on agentic coding, the more parallel sessions you end up with — and the way we operate them hasn't caught up. We're managing AI agents the same way you'd manage background processes in 2015: by tabbing through terminals and hoping you catch issues in time.

The insight is that an agent session isn't a script you fire and forget. It's a process with state, with context, with real-time decisions — and it deserves the same observability any production system gets.

So we built the layer that sits between you and your agents. Agent Cockpit is a local daemon that intercepts every tool call before it executes (via the hook lifecycle), streams session state in real time over WebSocket, and gives you one dashboard over all of it. Including an Office Map View that turns each session into a character on a 2D canvas — sounds gimmicky until you have 5+ agents running and a flat list stops working.

A few things that matter to us on this launch:

It's MIT. No telemetry, no paid tier hiding behind v0.2, no open-core trap. The orchestration layer for agents is too foundational to be locked behind a SaaS this early in the curve.

It's local-first. The daemon runs on your machine. Your agent traffic doesn't round-trip to a cloud we control.

We just shipped v0.1.4 on npm. It's the start — deeper CLI support, an oversight layer that uses a model to flag suspicious agent behavior, and cross-session coordination are all on the roadmap.

Would love to hear what you think — especially if you've been running multiple agents in parallel and felt this pain yourself.

Try it free at agent-cockpit.dev or just run this command in your terminal: npx @agentcockpit/agent-cockpit