PermeantOS moves sensitive AI agent state, including prompts, tool outputs, intermediate activations, and potentially references to credentials or private artifacts. Treat migration artifacts as sensitive by default.

PermeantOS is currently a research preview. Security fixes will target the main development branch until versioned releases are established.

Please report suspected vulnerabilities privately to the project maintainer before public disclosure. If no private security channel is configured yet, open a GitHub issue with minimal non-sensitive detail requesting a private disclosure path.

Do not include:

• Real user prompts or conversations.
• API keys or cloud credentials.
• Private model weights.
• Migration manifests containing sensitive context.
• Exploit payloads against third-party services.

• Encryption and signature envelope.
• Migration manifest contents.
• Runtime adapter hooks.
• Cloud provisioning scripts.
• SSH tunnel setup.
• Prefix-cache and KV block attachment logic.
• Future Agent Memory Graph export/import.

• Run cloud E2E tests in disposable environments.
• Verify cleanup of instances, security groups, keys, and local PEM files.
• Avoid exposing the daemon directly to the public internet.
• Prefer SSH tunnels or private networking for remote validation.
• Redact manifests before sharing when they contain real prompts or model state.