Start
AI Insights docs
What AI Insights does with your local coding-agent history, what you get from it, and where to go next.
On this page
AI Insights reads the history your AI coding agents already keep on your computer (Claude Code, Codex, Cursor, Pi, OpenCode and Gemini CLI) and turns it into one HTML page: a recap of animated cards you can post, and a private analysis of what to change. It runs on your machine, makes no network calls and never calls a model API.
npx ai-insights@latestThat is the whole install. You need Node.js 20 or newer. The rest of these docs explain what the page shows, what the tool reads, and the optional parts: an agent review that adds judgement to the numbers, and a public leaderboard you can join from the website.
What you get
- A recap to share. Up to 25 animated cards about how you work, in three styles and four shapes, with video and image export. Cards hold counts and a few allowlisted words, never your prompts, paths or project names. See Recap cards.
- A private analysis. Your rhythm, parallel sessions on a timeline, tokens and an estimated cost at API list prices, findings with the evidence behind them, and the changes ranked, each with text you can paste into
CLAUDE.mdorAGENTS.md. See The report. - Change over time. Every run saves a small snapshot. The next run shows what changed: findings that look resolved, still open, or new.

You can open the sample report to click around before you run it on your own history.
How it works
- Read. AI Insights finds each agent's transcripts and parses them into sessions: numbers, tool metadata and the prompts you typed, with secrets redacted. It does not keep the agent's replies or tool output.
- Count. From those sessions it computes activity, the recap, token usage, cost and findings. No model is involved, so the same history always gives the same result.
- Write. It writes one self-contained HTML file to
~/.ai-insights/reports/report.html. The page loads nothing from the network. - Optionally, review. From inside your agent, the
analyze-ai-usageskill has your own agent read short, redacted excerpts of a sample of sessions, label each one from a fixed list, and write a short narrative. The next report includes it. See The agent review.
What it is not
- Not a monitor. It reads files after the fact. It does not watch your agents or run in the background.
- Not a bill. Cost is an estimate at API list prices. If you pay through a subscription, your real spend is different.
- Not a backup. It can only look back as far as your agents keep their files, and it keeps the stats of deleted sessions, not the transcripts. See How long agents keep history.
- Not verified. Leaderboard numbers are self-reported. The site checks the format of what you publish, not that it is true.