Your agents did something last night.
Can you prove what, and who approved it?
genevalakes Fleet is the accountability layer for autonomous agents — identity that can’t be borrowed, approvals that hold, and a ledger that won’t quietly lose the one finding you needed. Installed on your machines, running on your own AI account.
Anyone can run agents. Almost nobody can account for them.
The hard part was never getting an agent to do the work. It’s the morning after — when something expensive happened and the only record is a chat log nobody can search, signed by nobody, approved by no one in particular.
- An agent acted on a claimed approval that never came from a human.
- A tool reported “all clear” from a check that never ran.
- A finding got overwritten and nobody noticed for a week.
We know those three happen, because each one happened to us — and each one is why the corresponding control exists.
Three controls. Each one is a scar.
1 Identity that can’t be borrowed
Every agent carries a serial and signs its own work. An agent handed a claimed approval from another agent refuses to act on it — including from the fleet manager. Approvals come from you or they don’t count.
2 Gates that hold under pressure
A denied action stays denied. When an agent finds a second route to the same change, it refuses to take it and says so. That’s the behaviour you’re buying.
3 Agents that correct themselves
When a measurement contradicts an earlier finding, the agent supersedes its own work and leaves the original standing as the record of what was believed. That’s the difference between a confident answer and a trustworthy one.
Running in production, not a prototype.
- agents under management
- 32 agents under management
- machines under the conductor
- 12 machines under the conductor
- findings recorded
- 750+ findings recorded (639 currently open)
Measured on our own fleet, 24 September 2026.
Your AI account. Your bill. Your ceiling.
You hold your own Anthropic or xAI credential. We install and operate the governance layer against it — your model spend is billed to you, directly, by them. We never touch it, never mark it up, and never become a line item between you and your vendor.
The part that pays for itself: the same fleet, configured badly, can cost twenty times what it costs configured well. Caching, batching and routing routine work to cheaper models is part of the install. We’re the ones who bring that bill down — we have no reason to want it high.
And you’re not locked to one vendor. Our launcher runs agents on either of two AI vendors, and we have completed production sessions on both — not a config flag, actual finished work. If your provider changes their pricing or their terms, that’s a configuration change, not a migration.
Priced per fleet, not per agent.
Charging per agent would punish you for using it. Install equals one month, and it’s waived on an annual commitment.
Govern
Most commonUp to 8 machines
$5,000 /mo
Install equals one month, waived annually
Book a walkthroughAll tiers: identity-bound agents · approval gates with spend caps · the findings ledger, with export · phone-driven control · runs on either AI vendor.
Attest adds: an exportable audit trail built to be shown to a regulator, an insurer or a client.
Additional machines $250/mo each. More than 20 machines, or a regulated industry? That’s a conversation, not a checkbox — Let’s scope it.
The questions worth asking.
- Isn’t this just observability? I’ve seen tools at $40 a seat.
- Those trace what your app did so a developer can debug it. This decides what your agents are allowed to do, holds the approval, and keeps the record. A tracing tool has never once refused to take an action.
- Why can’t I just build this?
- You can. We did — over three months, across twelve machines, and every control on this page exists because something went wrong first. The build isn’t the hard part; knowing which controls matter is.
- What happens when my AI provider changes pricing?
- You change providers. Our launcher runs agents on either of two AI vendors and the governance layer doesn’t care which is underneath. That’s the whole reason it’s a separate layer.
- Who’s liable when an agent does something expensive?
- You set the spend ceiling and you approve the actions that need approving — that’s what the gates are for. Our liability is capped at fees paid and excludes anything you approved through a gate. The audit trail exists precisely so that question has an answer.
See it running on your fleet.
A walkthrough is a real session on real machines — not slides.
Book a walkthrough