Access control for AI agents on Postgres

Put your agents to work.Keep control.

Give each agent scoped access to your database. Permitted work runs. The writes you pick wait for someone on your team.

Evaluate a workflow with us See how it works

One agent workflow, in your staging environment. We set it up with you.

support-agent / orders dbexample

Approved dana@acme.testin the FaultWall app

Statements and outcomes from our v0.7.0 test run on a synthetic database.

In practice

Most of what an agent does should just run. You decide the exceptions.

Scope it

Each agent gets its own key and its own write rules, set in the FaultWall app.

Let it work

Reads and permitted writes go straight through to Postgres. No agent rewrite.

Decide the exceptions

Writes you selected wait for a reviewer. Blocked operations never reach the database.

Try it

Look up the order. Ask before changing it.

Say your support agent can read orders but needs a yes before it changes one. It does the research on its own. The change waits at FaultWall, before it reaches Postgres.

What happens to each statement

support-agent / orders dbexample

SELECT id, status FROM orders WHERE id = ?ALLOW

support-agent wants to runUPDATE orders SET status = ? WHERE id = ?

Order 1043 in Postgres processing→refunded

Held at the proxy. Nothing has changed in the database.

Work with us

Bring us one workflow.

Tell us what your agent should be doing that it isn't allowed to do today. We'll put it under limits your team agrees on, run it in staging, and give the owner the evidence to decide on production.

Evaluate a workflow with us See how it works

One agent workflow, in your staging environment. We set it up with you.

Trusted by engineers at

pluto