Last updated September 2026
Privacy
Poggle stores what your coding agents do. That is unusually sensitive data, so this page is specific about what is kept, what is stripped, and what never arrives at all.
What we store
Session metadata (which agent, which plan label, which branch, when it started and how it ended), an append-only log of events with short summaries and structured detail, and — when you enable it — segmented transcripts of the agent conversation.
Account data is limited to what is needed to sign you in and bill you: an email address, workspace membership, and subscription state.
What is stripped before storage
Credentials are scrubbed on your machine before anything is sent, and scrubbed again on arrival. Bearer tokens, provider API keys, JWTs, connection-string passwords, private key blocks and sensitive environment assignments are replaced with markers.
Home directories are removed from file paths, because an absolute path identifies a person and a machine while a relative one does not.
Bulk tool output is truncated at capture. We keep the head and the tail; the middle is dropped and not recoverable.
What we do not do
Your session data is not used to train models. It is not sold, and it is not shared with third parties beyond the infrastructure providers needed to run the service.
We do not read your logs except when you ask us to help with a specific problem, and we would rather you redact and paste than grant us access.
Retention and deletion
History retention follows your plan. Deleting a project removes its sessions, events, transcripts and briefs. Deleting a workspace removes everything belonging to it.
Revoking a relay key stops new data arriving from that machine immediately; it does not remove what was already logged, because the log is append-only by design.
Self-hosting
Poggle is a Next.js application over a Postgres database. If none of the above is acceptable for your code, run it yourself and this page stops being about us.