Run AI your auditors can check.

Live demonstration

Try to break the audit trail.

This is the chain of custody behind every governed action: envelopes signed, gates decided, entries hashed and sealed. Watch the evidence arrive, then tamper with a block yourself and see the verifier catch the break at the exact position.

Live proof

We run our own diligence on it.

When investors review Quox, the dataroom they use runs on the same protocols the platform ships. Every login, document view and export becomes an AEE envelope, a VOLT ledger entry and a WARD-witnessed receipt, and the session’s trail is theirs to inspect and verify from inside the room itself.

The same pattern is live on this site: quox.ai/facts.json carries a WARD receipt you can verify offline, with no Quox account. Read how it works →

Security controls

Controls your security team will recognise.

Governance that holds in production: every gate decision is logged, and a step that crosses a line you drew waits for a human.

At scale

Govern AI agents at scale, not one agent at a time.

As the number of agents you run grows from one to an estate, the question stops being “is this agent safe” and becomes “is every agent in this organisation held to the same standard”.

AOCL’s policy gate is the same layer for every agent, so a policy written once, blocking a dangerous command or requiring approval above a threshold, applies to the whole estate: a new agent added to the organisation inherits that policy set from day one, rather than starting ungoverned until someone remembers to wire it up.

Fleet products from the big clouds watch fleets that live in their clouds. Quox governs from your side of the boundary: one self-hosted control plane, tied to no model vendor, with evidence you can hand to an auditor.

Baselines are tracked per agent, so drift in one agent’s behaviour shows up against its own history, not a fleet average.

And when containment is what the moment calls for, E-STOP halts an individual agent and DEFCON RED steps the whole instance down: enforced server-side at every dispatch chokepoint, human-only to set or release, failing closed, with the trigger itself landing in the same audit trail as everything else. Alongside them: the approval hold, the policy deny, the budget gate and the loop-stuck kill.

The layer above

The questions a runtime cannot answer.

Runtime guardrails answer one question: was this tool call permitted. Quox models the organisation, not just the agent: objectives that give agents a reason to run, teams and approvals that give them authority, budgets that bound them, and an evidence chain that outlives them. Self-hosted, tied to no model vendor, on infrastructure you control.

Compliance

Mapped to the frameworks your auditors ask for.

One evidence chain, exported against seven frameworks. WARD-signed, generated from live trace data, not screenshots and spreadsheets. Each export fills the sections your telemetry can evidence and marks the rest action-required.

The big products

Six products, one evidence chain.

Each ships as its own thing. Together, they are how the claims on this page actually run.