Documentation
Execution-Layer Security (ELS) for AI agents — out-of-band command approval for Kimi Code CLI and Claude Code.
Beta documentation
These pages track the current Beta pre-release
(6.1.0bx). Behavior may change between pre-releases. The
authoritative long-form documents live in the
repository.
What AIGuard is
AIGuard is an Execution-Layer Security tool: it secures the layer where an agent's decisions become real actions — process and command execution on your machine. It hooks into your assistant's official PreToolUse mechanism, evaluates every shell command against a policy, and returns one of three actions: allow ask block
It is a guardrail against honest-but-fallible agents — not a sandbox, and not a defense against adversarial ones. See Risk Disclosures before relying on it.
Quick start
pipx install aiguard-gate
aiguard setup
aiguard install kimi # or: aiguard install claude
Full instructions: Installation.
Pages in these docs
| Page | Contents |
|---|---|
| Architecture | Evaluation flow, policy engine, approval flow, state layout |
| Installation | Requirements, pipx install, TOTP enrollment, hook setup |
| Releases | Versioning, release process, support policy |
| Risk Disclosures | Operational warnings — read before relying on AIGuard |
| Security | Guarantees, known limitations, vulnerability reporting |
| Contributing | Development setup, workflow, design records |