Security overview

A personal assistant stores your API keys, conversation history, credentials, and memory. GenosOS treats security as a first-class architectural concern, not a configuration checkbox.

​

Two-layer security architecture

GenosOS uses two distinct layers that serve different purposes and have different trust properties.

​

Immutable layer

Defined in ## Safety in system-prompt.js at position 4. This layer enforces anti-injection patterns, identity verification, and session integrity. It is hardcoded in the codebase — no agent, user, or external content can modify or delete these rules. This is architectural protection, not just textual policy.

​

Personalizable layer

Defined in SECURITY.md in your workspace at position 18. This layer covers scope of trust, vault awareness, channel restrictions, and custom red lines. It is editable per-agent and per-workspace, giving you control over policies while leaving the immutable foundation intact.

​

Security layers

​

Gateway binding

The gateway binds to loopback only (127.0.0.1) by default. It refuses to start on a non-loopback interface without an auth token.

​

DM pairing

The default DM policy is pairing: unknown senders receive an 8-character alphanumeric code and their message is not processed until you approve it. Public access requires explicit opt-in (dmPolicy="open" plus "*" in the channel allowlist).

​

Channel tool restrictions

Not all channels carry the same trust level. GenosOS applies a deny-only pipeline step automatically:

• WebUI — unrestricted (localhost + WebAuthn already protect this surface)
• WhatsApp / Telegram / Discord / Signal — exec, bash, and process commands are denied
• Voice calls — nearly everything denied except session management

This means a WhatsApp contact cannot instruct the agent to run shell commands, even if the agent is otherwise fully capable.

​

Tamper-evident audit log

All security-relevant events are written to an audit log protected by HMAC checksums. Any modification to a log entry is detectable.

​

Automated security checks

Run genosos doctor at any time to surface risky or misconfigured policies. The autonomous doctor engine runs 7 checks and can auto-fix common issues. You can also ask your assistant directly: “run security audit”.

​

Next steps