Who changed what, always.

Agents act inside the same walls as people, and nothing they do is hidden.

Each workspace is walled off in the database: Row level security is forced on every table that holds your records, and a test fails if a new table ships without its policy.
Your AI signs in as you: Clients connect with OAuth 2.1 and PKCE. An agent sees only what its person can see, and takes no extra rights.
Every change has a name on it: Each edit records the person or the agent that made it, and most edits can be undone for 30 days. A sent email or a signed document cannot.
Some things stay with people: An agent cannot decide an approval, change roles or billing, export your data or delete a workspace.
An audit log you can read: Sign-ins, role changes, exports, API keys, agent sessions and pay reads are logged. Shares show in each record's history.
Sensitive fields are encrypted twice: ID numbers and health values are sealed with a key per workspace on top of disk encryption. Mailbox tokens are sealed with AES-256-GCM, and API keys are stored only as hashes.
Hosted in Singapore: Agos runs on Google Cloud in asia-southeast1. The database is backed up daily with point-in-time recovery, and secrets sit in Secret Manager.
Sign-in on your terms: Limit a workspace to your email domain and require Google sign-in for every member. Guests sign in with their own email.
Your data leaves when you say: An owner can export everything at any time. A deleted workspace is held for 30 days, then purged, and you get a certificate with the row counts. Backups clear within 35 days.
Locked down in the browser: Every page is served over HTTPS only, cannot be framed by another site, and loads scripts from Agos alone.
Files are never public: Uploads sit in a private bucket that refuses public links, and every overwrite keeps the old version.
Limits on every caller: Each session, person and workspace has a cap on reads and writes a minute, so a runaway script or agent slows down before it does damage.