Questions, answered.

Yes — free forever for developers, and open source under Apache-2.0. There's no paid tier gating your own profile, and no account required to run the assessment.

Your AI coding sessions that already sit on your machine — Claude Code, Cursor, and Codex — plus your local git history. It derives signals about how you work (prompting, orchestration, recovery), and it never reads your source code into your profile.

No. The assessment runs entirely offline from local files — there's no server in the path. Nothing leaves your machine unless you explicitly run publish, which shares derived signals only, never your code or prompts, and is revocable any time.

It measures patterns and signals from how you build, never the source itself — so your strongest, least-shareable work becomes provable without exposing a single line.

Claude Code, Cursor, and Codex sessions today, plus your local git history. The reader is adapter-based and open source, so support for more tools grows with the community.

One clone and one command: git clone https://github.com/nextmillionai/nextmillionai && cd nextmillionai && python3 -m nextmillionai start. It runs on Python 3.9+ — no Docker, no account, no signup.

No. The whole assessment is computed locally from files already on disk, so it works fully offline. The only time anything touches the network is if you choose to publish a derived profile.

Only derived signals — your scores, archetypes, and positioning — never your code, prompts, or files. It's explicit (it never happens silently) and revocable: you can take a published profile down any time.

Plain arithmetic over counted signals from your sessions, against research-anchored bands. No model ever writes your number, the methodology is versioned and public, and anything unmeasurable is marked insufficient, never estimated.

No — and it never will. There are no percentiles, cohorts, leaderboards, or “top X%”. Your positioning is a map of how you build, not a ladder. Different builder kinds are crafts, not rungs.

It's open. Paxel shipped closed, so no one could see it uploading data until the community caught it. Here the scoring, the methodology, and exactly what leaves your machine are all there to read — you don't have to trust a pitch, you can read the code.

No. A model is used only to write the narrative of your own profile — it never assigns a score, never sees other people's data, and nothing of yours is used to train anything.

macOS and Linux, on Python 3.9 or newer. It's a local command-line tool with an optional local web view — it doesn't need Docker or any cloud service.

Apache-2.0 — permissive and free forever. It's built by nokk labs and shaped in the open: the code, the scoring, and the methodology are all community-contributable on GitHub.