The compute plane

tie-runner is optional. It turns content-addressed trees into a compute substrate: a job takes an input hash, runs in a container, and produces an output hash.

tie-run run ghcr.io/you/tool:1 <input-hash> <output-hash>
tie-run status
tie-run jobs
tie-run logs <job>
tie-run deploy --prod <image>

The runner has its own HTTP API (port 1163) with jobs, pipelines, services, capacity, a queue and cancellation. tie-exec is the container entrypoint: it reads inputs and writes outputs with #!TIE markers on stdout. tie-fleet is a thin admin client for fleets of runners (brandable via fleetcmd.Run).

The normative protocol is in docs/runner-api.md; the wire fixtures are in testdata/runner/.

This is the most opinionated part of tie and the least necessary: if you want a library, you never need to start a runner. If you want reproducible derivatives — thumbnails, transcodes, OCR, ML inference — the hash-in/hash-out contract is the point.