Let AI agents into your ISMS without handing over the keys
AI agents have quietly become good at the kind of work an ISMS is full of: drafting a report, walking through a procedure, updating a register, checking what a policy actually says. So the obvious next step is to let one of them do it. The obvious next worry is what, exactly, you've just handed it access to.
Most integrations answer that worry badly. They mint an API key with its own sprawling permissions, or a service account that quietly outranks the people it works alongside: a second, invisible set of keys to your compliance system. Aquil's API and MCP service is built on the opposite principle: a connected agent is never more than the person it acts for.
The connection is you, and nothing more
When you connect an AI client to Aquil, the connection authenticates who you are. It does not carry permissions of its own. Every single request re-derives what the agent may do from your live role, your scopes, your organization and your billing status: the same checks the app runs when you click a button yourself.
The consequences are worth spelling out. An agent acting for a read-only user can only read. It can never reach another tenant's data, because your organization is resolved on every call and never taken from the request. And the moment your access changes (a scope removed, your account disabled, the add-on switched off, a token revoked) the agent's access changes with it, on the very next call. There is no long-lived grant sitting in a config file that outlives your permissions.
Connecting is a paste and a click
Because Aquil's MCP endpoint is also a standards-compliant OAuth 2.1 server, connecting a modern client is a paste and a click. In Claude, Cursor, VS Code or ChatGPT you add a custom connector, paste your Aquil server URL, sign in with your normal Aquil login, and approve a consent screen that lists exactly what the connector will be able to do. No token to copy into a file, no bridge script to babysit. The connected app then appears under Settings → API & MCP, where you or an administrator can disconnect it at any time, and the disconnect takes effect on its next request.
For headless uses (a CI job, an automated agent, a client without a connector dialog) you can mint a bearer token instead. Tokens are shown once and stored only as a hash, last six months, warn you in-app before they expire, and can be narrowed to a subset of capabilities. An administrator can revoke or renew any token; the owner can rotate it for a fresh secret.
What an agent can actually do
Access is organized into capability scopes, and an agent only ever sees the tools its scopes (and your role) allow. The tools span the ISMS: read policies, discover and fill reports, browse and add register entries, run a process step by step, and schedule audit reviews.
The design nudges agents toward doing the right thing. They discover before they write: they list a report's fields, required flags and valid options before filling it, so they can't invent a value that doesn't belong. Completed reports are never exposed, because they are immutable records. And when an agent fills a report it merges: it changes only the fields you asked for and leaves everything else exactly as it was.
Guardrails that match your own
The rules an agent runs into are the same ones a colleague would. If it doesn't own the register it's adding to, its entry is submitted for the owner's approval instead of being written directly, precisely as it would be for any teammate in that role. It can delete only a draft report it created itself through the API, never one a human made. A run flagged as confidential stays confidential. It can't file a report under a scope you don't administer.
And everything it does is on the record. Every action is written to your organization's Activity log, attributed to you, with the connecting token noted, so "the agent did it" is never the end of the audit trail. It is the start of one. Calls are rate-limited per user and per organization, so a runaway loop can't stampede your data.
The promise of an AI agent in your compliance workflow is real: the tedious parts (the filling in, the walking through, the looking up) handled for you. The catch has always been access. Aquil's answer is to make the agent a faithful stand-in for you and never a superset of you, so you can hand it the busywork without handing over the keys. It's an add-on under Settings → API & MCP, and on the Free plan it's switched on out of the box.