Skip to content
Axiom is live in early access, by invite.Join the waitlist

Security

Clear boundaries. Real control.

How Axiom handles approvals, spending, checked work, workspace isolation and outside content. Each control, and where it stops.

Actions that matter wait for you.

Before a Role takes an action that can't be undone, or reaches for a tool it hasn't been given, it stops and puts the request in Approvals, with its reason. Some actions, like publishing outside your workspace, always need a person.

Anything outside a Role's reach is refused outright, not asked. Changes inside what you've already approved can go ahead without you.

Three limits on spend.

Work pauses when spend reaches your funds, your optional monthly or total stop-at, or a pace limit you can't change, which guards against sudden bursts.

Planning is paid for too, and starts before any work is delivered. The check runs every minute, so spend can pass the limit slightly.

A separate check on the work.

Nothing a Role hands in counts until a separate checker has passed it.

A pass isn't a guarantee. Review the work before you use it.

Each workspace is fenced off.

Every piece of a workspace's data is fenced by row-level security on its workspace: with no workspace set, a query sees nothing.

Workspaces share one database. The separation is enforced in the database, not by a separate environment for each workspace.

Outside content stays data.

Content from outside, including what you write, reaches a Role only as labelled data, never as instructions. Outside content never becomes validated knowledge.

Labelling lowers the risk of text that tries to steer a Role. It doesn't remove it, so review work built from outside material.