[feat]: OrcaRouter provider support for Aeon

[armyluki-wq]

Aeon solves a problem most agent frameworks punt on: work that has to finish while you're away. A skill is a single SKILL.md plus a schedule in aeon.yml, and the self-healing loop — heartbeat → skill-health → skill-repair → self-improve — is what makes going unattended safe: a model grades every run and repairs broken skills instead of waking you at 3am.

The nine-harness contract is the design choice I keep coming back to. The same SKILL.md runs on Claude, Grok, Codex, Pi, Vibe, Kimi, fx, Cursor, and Hermes behind one run-harness contract, with the same result/usage/session shape. Anyone who chose Aeon for that kind of engine independence is exactly the user who wants a say in where inference comes from.

I'm an engineer on the OrcaRouter team, and I'd like to propose adding OrcaRouter as an optional gateway.

Proposal

Add OrcaRouter as an optional LLM gateway alongside the existing ones. It would not replace or change any current provider — existing native and sidecar gateways keep working exactly as they do today.

OrcaRouter exposes an OpenAI-compatible API at api.orcarouter.ai/v1 and uses standard API-key authentication (ORCAROUTER_API_KEY). That places it in Aeon's sidecar tier as described in CONTRIBUTING.md: an OpenAI-compatible provider bridged per run by the claude-code-router sidecar — the same shape as Venice and Surplus, a start_ccr_sidecar <slug> <openai-url> <key> <model> call plus a gateway-registry entry and a secret row. I haven't written or tested any code; this is a proposal, and I'd follow the documented five-file gateway pattern if you want it.

What it could add for Aeon users

Since Aeon already tracks per-run cost in memory/token-usage.csv and isolates billing per spawned instance, a few OrcaRouter capabilities line up with that:

• Automatic model routing and provider failover — fallback chains that keep an unattended cron run alive when one upstream degrades, which matters most when nobody is watching.
• Prompt caching — cached tokens billed at the provider's cache rate, useful for the repeated prompt prefixes in scheduled skills.
• Budgets and usage tracking — spend caps and per-request logs with model, latency, and cost, complementing Aeon's own token accounting.

Ecosystem

OrcaRouter is already used across open-source projects including goose, promptfoo, Dify, and RAGFlow.

Disclosure

OrcaRouter runs an optional open-source partner program: approved OSS projects can receive a 5% revenue share from OrcaRouter usage attributed to their integration. Participating is not a prerequisite for the integration, and I'm happy to follow whatever disclosure or governance requirements this project has.

More detail: https://www.orcarouter.ai/built-with

Would you be open to this? I'd welcome your thoughts on whether the sidecar tier is the right fit, and I'm glad to open the implementation PR if you'd like one.

[aaronjmars]

Thanks for the thorough writeup, and for the upfront disclosure on the partner program.

The sidecar tier is the right fit - an OpenAI-compatible gateway bridged per run via claude-code-router, the same shape as Venice and Surplus, is exactly the five-file gateway pattern in CONTRIBUTING.md. We are open to it as an optional provider that does not touch any existing gateway, so please go ahead and open the implementation PR following that pattern.

A few asks for the PR:

• Keep it fully optional and off by default; existing native and sidecar gateways stay unchanged.
• Wire ORCAROUTER_API_KEY as a normal secret row and register it in the gateway registry per the documented pattern.
• Note the revenue-share arrangement in the PR description for transparency.