Security
Last updated June 15, 2026.
Your code stays on your machine
The agent runs locally in your working directory using your own Claude subscription. We never receive your source code or your Claude credentials. What leaves your machine is the session transcript you choose to share, so your team can follow along.
Secrets are scrubbed before they leave
Agent output is scanned and known secret patterns are redacted before the transcript is mirrored to the web, so credentials in command output are not broadcast to the room or stored.
Credentials are stored as hashes
A session’s host key and its orchestrator control token are high-entropy random values, and only their SHA-256 hashes are stored. The raw values exist only on your machine and, for a control link, in the URL fragment, which is never sent to a server or written to a server log. Control tokens are revoked the moment a session ends.
Access control
Prompts from the web queue for host approval by default; the host approves what runs, can trust specific participants, and can lock posting to themselves at any time. Authentication for accounts and organizations is handled by a dedicated third-party identity provider.
Transport & hardening
All traffic is over HTTPS and secure WebSockets. The app ships a strict Content-Security-Policy and standard hardening headers (no framing, no MIME sniffing, a referrer policy), and treats all participant-submitted content as untrusted.
Reporting a vulnerability
Found something? We want to know. Contact us with the details and we will follow up.