Bring your own keys
Use your own Anthropic or OpenAI keys; pay your provider directly.
BYO keys lets you bring inference billing in-house. The harness uses your key for the model call; the catalog doesn't see the response, only the routing decision and the structured fields the workflow extracted.
Supported providers
- Anthropic — any model the workflow's YAML declares as an acceptable target.
- OpenAI — same constraint.
- Custom providers via OpenAI-compatible endpoints are queued; see /roadmap.
Key storage
Keys are stored encrypted at rest, scoped per-subscriber, and never written to logs. They are used only for the model call the harness makes during a run. Rotating a key takes effect on the next run; in-flight runs complete with the previous key.
Billing implications
BYO keys removes the model-cost share from your Pro subscription — Pro stays $19/mo and the inference invoice lands at your provider. Run-history retention halves to forty-five days as a side effect: the response itself is never stored, so the harness has less of the run to remember.