Route
Every model call through one local-first router — tiers, fallbacks, and cost control across 19+ models. No cloud dependency, no lock-in.
The control plane for AI agent fleets
Your agents did something. Prove it.
Teams past the demo stage hand-roll the same four things: model routing, fleet coordination, an audit trail, and a durable record of every session. FleetOpus fuses them into one operable surface — and writes a witnessed receipt for every action.
Route it. Run it. Prove it.
Live chart — fleet-mr4x48nr
witness ledger · awaiting gated actions…
Run
Spawn and coordinate agent fleets with quorum decisions and per-recipient receipts — process-isolated, framework-agnostic, attachable to the tools you already use.
Prove
Every message witnessed, every consumption receipted. When someone asks “what did the agents do and who approved it?” — you answer with evidence, not vibes.
An agent asked to delete something. Here's the receipt.
Real output from a FleetOpus run: a 3-agent fleet merges some notes, and one agent requests a destructive delete. The call blocks until a quorum of its peers ratifies it on the witness bus — then the trail answers who saw it, who approved it, and why.
Witness ledger — entry № 18,404replayable
Silence times out to deny. Self-votes don't count. Nothing outside the whitelist runs. Every line above is a receipted message you can replay.
Provenance
Built from operational history, not a whiteboard.
The coordination design behind FleetOpus ran a 10+ agent fleet in production for 40 days — quorum-ratified decisions, self-retraction chains, all on the record.
- Days at sea
- 40
- Witnessed msgs
- 18,404
- Unexplained
- 0