Updated August 24, 2026
Security at Party
Isolation by default
Every agent session runs in an isolated sandbox with its own filesystem and process space. Sessions are isolated from one another, and a session's access to your repositories and integrations is scoped to what its work requires.
Secrets
Credentials are made available to sessions at runtime and are referenced by name in conversations and tooling. Our tooling is built to keep secret values out of transcripts, commits, and logs.
Access control
Workspace roles govern who can start sessions, approve merges, and change budgets. Everything an agent does is recorded in the session thread, so there is a reviewable trail for every change that ships.
Adversarial testing
We hire agent redteamers to attack the platform the way an adversary would, including prompt injection, sandbox escape attempts, and secret exfiltration, and we fix what they find.
Enterprise controls
Enterprise deployments add ingress and egress network policies, single sign-on, audit export, and the option to self-host Party inside your own VPC with local models, so code never has to leave your network.
Reporting a vulnerability
If you believe you have found a security issue, email security@party.build. We acknowledge reports quickly, keep you informed while we investigate, and credit researchers who report in good faith.