Security

The centre of Okestri’s design is that an agent stops before it does anything that matters. Sending an email, changing your code, spending money — each of these pauses and waits for a person to approve it. This is not a setting you switch on; it is how the system is built, and it applies to every tool an agent can reach, including ones you connect yourself.

You choose which routine actions may happen unattended. Anything outside that list stops and asks. Anything you have explicitly forbidden is refused outright and never even reaches you for approval.

Each job gets its own isolated workspace with a private, shallow copy of the code it needs. Work happens there, not on your machines. Commands are stopped at a time limit and their output is capped, so a job cannot run away.

When credentials appear in command output, they are removed before that output is stored or shown.

One thing we want you to hear from us rather than discover: a workspace is not cleared automatically when a job ends. The copy of your repository stays on our infrastructure until you reset or delete that environment from the Environments page. Automatic expiry is on our list. Until it ships, Environments is where you clear anything you would rather not leave sitting there.

An approval that sits unanswered does not wait forever and does not quietly go ahead. It reminds once, and if it is still unanswered when it expires it is declined automatically and nothing is executed. A job started at two in the morning cannot hold a live action open indefinitely, and it cannot proceed simply because nobody was awake to look at it.

Where a service supports it, you connect by approving Okestri on that service’s own site — we never see or store your password. The resulting access is encrypted before it is stored, and can be revoked from Okestri or from the service itself at any time.

Credentials are never given to the AI model. An agent asks Okestri to perform an action; Okestri performs it using the stored access. The model itself never sees the key.

Model provider keys are yours: you supply your own Anthropic, OpenAI, or Google key, it is encrypted at rest like any other credential, and you can revoke it with that provider without going through us.

Passwords are stored hashed, never in readable form. Two-factor authentication is available and we recommend turning it on. You can see every active session on your account and sign any of them out. Repeated failed sign-ins are rate limited.

Every job keeps a full record: what the agent examined, what it proposed, what you approved or declined, and what changed as a result. Approvals are recorded against the person who made them. This exists so that a disputed action can always be traced back to a decision.

We would rather tell you the truth than imply a badge we have not earned:

SOC 2

Not certified. No audit is currently in progress.

ISO 27001

Not certified.

HIPAA

Not supported. Do not put protected health information into Okestri.

GDPR

We follow its principles — see the privacy page for access, export, and deletion. We have not completed a formal audit.

Penetration test

No third-party test has been carried out yet.

Data residency

No choice of region today. Data is stored where our infrastructure runs.

Workspace expiry

Not automatic yet. Repository copies persist until you clear them from Environments.

If your company needs any of these before it can use a tool like this, tell us and we will be straight with you about whether and when we expect to get there.

If you believe you have found a vulnerability, email [email protected]. Tell us what you found and how to reproduce it. We will confirm we have received it, and we will not pursue anyone who reports a genuine issue in good faith.