Relay

A local harness for AI coding agents that manages runs, retains learning, and verifies changes before pull requests.

Relay runs AI coding agents on your machine with models you already use or your own API keys. It queues coding tasks, preserves run learnings, enforces budgets, recovers failed steps, and runs configured tests, lint, or builds before work is considered complete. Paid plans add unattended multi-repository runs and sandboxed checks.

What Relay does

• Runs coding agents locally with connected model providers or API keys
• Queues tasks and turns completed work into pull requests
• Retains learnings from runs for subsequent tasks
• Enforces turn, token, time, and spend budgets
• Runs configured tests, lint, and build commands as completion gates
• Re-runs checks in fresh sandboxes on paid plans

Visit Relay website