Security
The concrete mechanisms behind PawOS's safety and security claims.
Encryption
Organization-shared infrastructure credentials are stored in an encrypted vault, never in plain text. Data in transit to Supabase and third-party AI providers uses standard TLS encryption.
Authentication
Individual accounts use Supabase Auth (email or Google OAuth). Organization membership is verified against a shared email domain before access is granted.
Permissions
Access within an Organization Workspace is role-based, with temporary elevated permissions scoped to a specific task rather than permanent role changes.
Runtime isolation
Each runtime (Universal Execution, Browser, Infrastructure, Communication, Companion, Governance) operates through its own plugin contract, limiting how far a failure or misuse in one area can reach into another.
Secrets & credential storage
Individual-use infrastructure connectors rely on your own already-authenticated CLI/API sessions rather than PawOS storing your cloud credentials. Organization-shared secrets go through the encrypted credential vault.
Data handling
Most data stays local to your device by default. See our Privacy Policy and Data Processing Agreement for full detail.
Local execution
Universal Execution Runtime actions run on your own machine, under the same OS-level permissions your user account already has — PawOS does not escalate privileges.
Cloud providers
Every hosting/cloud connector shells to that provider's own official CLI or API using your own authenticated session. PawOS never manages your cloud provider credentials on your behalf.
Responsible AI
See our Responsible AI Usage policy for our commitments around honest task reporting and your responsibilities when reviewing AI-proposed actions.
Safety checks
Actions are classified as routine, destructive, or production-impacting before they run, with confirmation or organization approval required for the latter two.
Execution approval
Organizations can configure approval policies requiring a human sign-off before specific action types (like production deploys) run, backed by a real approval queue.
Autonomous Engineering safety
Autonomous Ticket Resolution's deploy step (where applicable) goes through the same approval gating as a manual deploy request, and billing is gated on genuine completion, never on partial or failed work.
Incident reporting
See our Vulnerability Disclosure Policy for how to responsibly report a security issue.
Security best practices
Authenticate infrastructure CLIs with least-privilege credentials where your provider supports it, scope organization approval policies to genuinely risky actions, and review confirmation prompts before approving them.