AI work orchestrator
Drop your whole build list into a single queue. Relay keeps every endpoint — your cloud subscription and your local machines — running flat out, at the lowest cost, automatically.
The problem
AI work is bursty and manual. Jobs get fired one at a time, at whichever endpoint is top of mind. Meanwhile two expensive resources sit wasted.
Subscription capacity resets every cycle. Whatever you didn't use is gone — paid for, never run.
A fast Mac and a GPU box sit doing nothing all night while the cloud bill climbs for work they could do free.
Big jobs stall waiting for a human to notice them and pick an endpoint by hand.
The promise
Backfills idle cloud quota before it resets — same bill, more shipped.
Drains free work to local machines the moment they go idle.
Sends each job to the right brain by privacy, cost, and latency.
One board for throughput, spend, and whatever's stuck.
The proof
A priority queue plus pluggable workers — a cloud API worker and a local runner — each aware of its own rate limits and capacity. Jobs flow to the first free, eligible worker.
Finish the build list in a fraction of the wall-clock time — for the money you're already paying.