onlyai-pilot

Run OnlyAI Pilot's guardrail checks inside Claude Code, Codex, Cursor, or any terminal. Pilot flags known drift/bug classes (optimistic state, hydration mismatches, unsafe long-jobs, silent write failures, …) — with more review lenses (security, architecture, …) on the way — and returns actionable findings with fixes.

How it runs: the checks run on OnlyAI's servers. The CLI collects the relevant source files from your repo (limited to app / lib / components / src / pages, no node_modules, no .env, size-capped) and sends them over TLS. Your code is processed in memory to produce your findings and is never stored, logged, or used to train anything — it is discarded the moment your results return. Only anonymous, code-free failure patterns (which rule fired + the outcome) are retained to improve Pilot, and only if you leave fleet learning on.

Requires an active OnlyAI Pro plan. Zero dependencies (Node 18+).

Install

npm i -g onlyai-pilot

Use

pilot login <token>     # get a token at onlyaiapp.com/settings → "Pilot in your terminal"
pilot check             # audit the current repo
pilot config telemetry off   # opt out of anonymous fleet learning
pilot help

Tip: add a line to your project's CLAUDE.md (Claude Code) / AGENTS.md (Codex) — "before deploying, run pilot check and fix any blocking findings" — and your AI agent will run it for you.

Config is stored at ~/.onlyai/pilot.json. Override the API base with --api <url> or ONLYAI_API.

License

© OnlyAI (Academy of AI). Proprietary — all rights reserved; requires an active Pro plan. The guardrail rules, prompts, and catalogs run only on OnlyAI's servers and are not licensed to you. See LICENSE.