nah documentation

nah is a deterministic safety guard for coding agents. It blocks visible tool calls it can prove are disasters — deterministically, without an LLM — and leaves everything else to the runtime's normal approval flow. Start narrow and load only the topic needed for the current task.

• Start — install nah and guard one agent.
• Core concepts — effects, verdicts, guards, and trust.
• CLI guide — the complete human and machine command map.
• Configuration — shipped policy and trusted projects.
• Extending — build one-shot guard programs.
• Agent runtimes — supported integrations and their boundaries.
• Threat model — adversary, assumptions, and defense in depth.
• Security boundaries — what nah does and does not enforce.
• Architecture — contributor-oriented code navigation.
• Changelog — user-visible changes.

The focused pages under docs/ ship inside the binary; this index and the changelog remain repository-only:

nah docs
nah docs start

Runtime-specific topics use names such as runtime-codex and runtime-hermes. The website publishes a generated guard catalog from the compiled binary. nah docs guards renders the same built-in policy catalog and tested examples, plus local enablement state and custom guard status.