Relay

Coding agents are useful, but long-running tasks are fragile. A terminal disconnects, a process exits, or the session loses the context it needs to continue — and the work stops until someone notices.

I built Relay as a local harness that supervises coding agents while you are away. It keeps task state, restores interrupted sessions, follows the repository's conventions, and runs your configured checks before reporting back.

Relay also produces signed receipts that can be reopened and rechecked. They are evidence of what ran and what passed — not a claim that the code is automatically correct or ready to ship. Human review still decides that.

The goal is simple: make unattended agent work recoverable, visible, and auditable while keeping the code and toolchain on your machine.

I would especially value feedback from developers already giving agents tasks that run for hours: where do your runs fail today, and what evidence would you need before trusting the result?

https://relayevals.com/