Hey Product Hunt 👋 Codeep is an AI coding agent that runs on your machine — reads your whole project, runs commands, writes and verifies code. It launched here in June. Three things have happened since that I think are worth your time. Agent Studio. Build an agent visually: pin it to an exact model, hand it only the capabilities it needs — files but no shell, tests but no git — and scope it to specific projects. It saves as a portable Markdown file you can commit next to your code. The limits are real, not a prompt asking nicely. A Files-only agent has no shell to reach for, because the tool was never offered to the model. And every run now writes down what it touched — including the calls the boundary refused — so you can check the fence held instead of trusting that it did. Approval on your phone. An agent is useful right up to the moment it needs you. It hits a command it won't run without permission, stops, and waits — correct behaviour, and also the moment everything falls apart if you walked away. The obvious fix is a server: run the agent centrally, approve remotely. I don't want to build that and I doubt you'd want to trust it. So Codeep does the opposite — the agent stays on your machine, only the decision travels. The question arrives over Telegram. No server, no inbound port, same on macOS, Linux and Windows. A redesigned Mac app. Rebuilt around the run itself rather than the chat log — you can see what the agent is doing, not just what it said. Free, any model (hosted or fully local), no account needed. Happy to answer anything — especially whether the phone approval fits how you actually work.

Codeep

Deep into Code.

Deep into Code.

Codeep 3.0

Previous Codeep Launches

Previous Codeep Launches