Agent Recap
A local companion for Claude and Codex

Make your next
session better.

Better agent work starts with the setup around it. Find the friction in your coding sessions. Make your next experiment count.

Try Agent RecapExplore a sample report

Free to use. Local reports. No account required.

A moment worth revisitingIllustrative example

The same request.
A second time.

Repeated instructions can point to something your agent setup is missing.

Try one small change

Capture the recurring procedure as a skill. Check whether it helps on a comparable task.

A coaching idea, not a measured result.

Your next improvement might be a clearer brief, a faster test, or one less repeated instruction.

Meet frontier engineering

Give every session a next step.

A useful report should lead somewhere. Our direction is a simple loop for improving how you work with agents.

Step 1

Observe

See recorded activity, repeated requests and missing evidence across your local sessions.

Available in Agent Recap
Step 2

Experiment

Review a session in context. Choose one change to your instructions, tools or workflow.

Optional AI review available
Step 3

Verify

Compare the next task with your expectations. Keep changes that help and revisit those that don’t.

Manual today; tracking is planned

A clearer view of your AI coding week.

Agent Recap reads local Claude and Codex logs and opens a report in your browser.

  • Filter by project, assistant and period.
  • Inspect evidence behind observations.
  • Export HTML, JSON and Markdown.
  • Choose an optional AI reviewer.
Read the usage guide
Start with synthetic sessions
npx --yes --package https://agent-recap.pages.dev/downloads/agent-recap-0.1.0.tgz agent-recap serve --demo --open

Node.js 22+. Downloaded from this site.
Remove --demo to read your own local sessions.

Bun and direct download options

Useful evidence. Clear limits.

Does it score my prompts?

Activity counts cannot establish prompt quality. Agent Recap shows observations and their coverage. Optional AI review gives contextual suggestions with turn references. Those are model judgments, not verified outcomes.

Do my sessions leave my computer?

Ordinary CLI reports are generated locally. AI review sends a bounded, redacted packet only after you explicitly choose to send it. You can use a compatible local model. This website does not receive your session logs.

Can I use npx agent-recap?

Use the exact installation command above for now. The short command will work once agent-recap is published on npm.

Find your next small improvement.

Run your first report