Security
Built so an agent can only do what its person can.
The agent has no permissions of its own. It borrows yours, for one request at a time, and asks before it acts.
What we do
Company-owned data
Conversations, memory, skills, and files belong to the company account, not to a personal login. Remove a member and their seat closes; the company keeps what it paid for.
Row-level security everywhere
Every table in the control plane is protected by row-level security keyed on the company and the person. The API never talks to the database with a caller's rights it did not verify first.
Short-lived tokens
The service mints a five-minute token per request after verifying your session. Browsers never hold a database credential.
Confirm before consequence
Sends, writes, purchases, and external actions pause for a confirmation. Background runs hold their confirmations for you to approve later.
Encrypted vault
Website logins used by computers are encrypted with a key that never leaves the service. Agents use a login without reading it.
Isolated computers
Each computer is its own Fly Machine with its own disk. Idle machines suspend, then stop. Nothing is shared between companies.
Optional two-factor
Admins can require TOTP for everyone in the company. Sign in with email, magic link, Google, or Microsoft.
Audit and export
Tool calls and usage are logged with who, when, and what. Export any month as CSV.
Where things run
The Agents API and computers run on Fly.io in the United States. Company data lives in a dedicated Supabase Postgres project in the same region. Payments are processed by Stripe; we never store card numbers. Model requests go to the vendor you chose, through OpenRouter or directly, and are not used to train models.
Questions, disclosures, or a security report: [email protected].
Give your team agents that answer to you.
Fourteen days free. No card. Invite your team in a minute.