It asks before it acts
- Sending, paying, deleting and publishing wait for your approval.
- An approval is bound to a digest of the exact action; change the action and it needs approving again.
Security
Autonomy you can leave running
TheOne is built so that an agent can do real work without being able to grant itself authority, grade its own results or reach data that is not yours.
Someone else checks the work
- Results are verified independently of the agent that produced them, with Ed25519-signed receipts.
- An agent reporting its own success is recorded only as a self-report; it does not close a verified task.
Your data stays yours
- Each workspace’s data is isolated by tenant in the database itself.
- Saved API keys are encrypted and used only for your own tasks.
- We do not sell your information or train our own models on your content.
- Staff do not browse content; any access, for support or abuse, is recorded with who, when and why.
Careful with your code
- Only the files a coding task was allowed to change are written back.
- Files that existed before a run and are not under version control are never touched.
- Bot webhooks are accepted only on each service’s own host, over HTTPS.
It cannot rewrite its own rules
- Self-evolution can never change its permissions, its verifier or its trust roots.
- Changes to source code always need a person to approve them.
Every release is audited
- CI and UI screenshot tests, then a candidate audit, a database restore drill and only then production.
- The release chain itself is locked: changing it takes an explicit amendment by the owner.
Found a security problem? Tell us through OneAI Labs; please do not disclose it publicly before we have fixed it.