An analysis is produced by your own AI coding agentrunning a frontier model against your actual repository — real files, real line numbers, real churn data, read with full agentic depth. It targets a strict schema, gets validated until it's clean, and only then is published here as an interactive onboarding site. The reading is done by the best model you have access to; the viewer just renders the result.
Run this at the root of the repository you want to onboard. It writes .repo-onboarding/ with the deterministic facts (file tree, LOC, languages, git churn), the agent protocol PROMPT.md, and the target schema.json. Nothing in your repo is modified — the pre-pass only reads.
npx repo-onboarding initHand this instruction to your AI coding agent. It reads the code and writes analysis.json at the repo root. Works with Claude Code, Cursor, Codex — any agent that can read files and follow a prompt.
Follow .repo-onboarding/PROMPT.md to produce analysis.json.Checks the document against the schema and verifies dependency-graph edge integrity. The errors are field-level and designed to be pasted straight back to your agent — repeat the validate/fix loop until it passes. This is the quality floor: nothing malformed gets published.
npx repo-onboarding validate analysis.jsonCreate a personal token at /account and set it as REPO_ONBOARDING_TOKEN (or pass --token). upload re-validates locally, then publishes and prints your live analysis URL.
export REPO_ONBOARDING_TOKEN=roa_…
npx repo-onboarding upload analysis.jsonOptional: commit an ONBOARDING.md
Want the guide to live in the repo too? export renders a validated analysis.json to a single committable ONBOARDING.md — the same content the hosted viewer shows, minus the interactivity. Purely optional; publishing below is the main path.
npx repo-onboarding export analysis.json > ONBOARDING.mdPrefer the browser?
Validate locally with step 3, then drag the file onto /upload to publish without the CLI. Same validation, same result.
Why no hosted generation?
Because the reading is the whole game. A frontier model with agentic access to your code beats any free model we could afford to host, and your source never has to leave your machine — you send us only the analysis.json you choose to publish.
What gets uploaded?
Only analysis.json — the structured onboarding document. Never your source code. You can read exactly what it contains before you run upload.
What does it cost?
Nothing here. You bring your own model, so generation runs on tooling you already pay for (or a free tier), and the hosted viewer is free.