Arcus

The Sovereign Gateway for AI Agents

Arcus is a security gateway for AI agents. Every request an agent makes passes through one control point that checks identity, policy, payload contents and behaviour before anything leaves your perimeter — and records the decision in a ledger that cannot be quietly edited.

The problem#

An AI agent is given an API key, network access and the authority to act on its own. Nothing in that arrangement supervises it.

The agent may be perfectly well-behaved and still cause the incident: it pastes a credential into an outbound message because the credential was in its context; it calls an endpoint nobody intended it to reach; it sends a DELETE where a GET was expected; it retries in a tight loop and floods a downstream service. None of that is a model failure. It is the absence of a control point between the agent and the outside world.

Arcus is that control point.

What Arcus does#

Agents stop calling the internet directly. They call Arcus, and Arcus calls the internet on their behalf — but only after a request survives five checks in sequence:

A request that survives all five is queued and delivered. A request that fails any of them is refused with a specific status code and a readable reason — and either way, the decision is appended to a tamper-evident audit chain.

Why a gateway, and not a library#

A library that an agent imports is a library the agent can be talked out of using. Arcus sits outside the agent process and holds the credentials the agent needs, so bypassing it means having no route at all. The policy that governs an agent is not part of the agent.

That also means Arcus works with any framework and any language. If it can make an HTTP request, it can be governed.

What you get#

Security is not a paid feature#

This is the deliberate commercial decision behind Arcus, and it is worth stating plainly:

Policy enforcement, DLP scanning, parameter guardrails, threat quarantine and the audit chain run on every request, on every plan — including free. Paid tiers buy more keys, more rules, deeper visibility and longer retention. They never buy the protection itself.

A product that puts the actual security behind an enterprise tier is selling you the absence of a vulnerability it chose to leave open. See Pricing & plans for exactly what each tier changes.

Where to go next#