An agent reads your sessions and tells you how to work better with agents · yours only
Once a week an agent reads your own coding-agent sessions — the whole transcript, not a count of them — and comes back with what you spent the week on and how to work better with your agents.
Each finding comes with a paragraph you can paste into CLAUDE.md, AGENTS.md or a knowledge doc, so your agents stop making the same mistake.
Your sessions grouped into real themes — not repo names — with the time and cost each one took.
Active, stalled or done, what the agent thinks the current state is, and what is still open.
A few tracked habits with a direction of travel, each backed by a quote from a real session.
This is the one feature that sends transcripts somewhere new — the Claude API — so it only runs if you say yes. Turn it on in AI analysis at the bottom of this page, or the next time the collector asks. Nothing has been sent until you do.
Whether an agent may read your sessions, and whose Anthropic account pays for it
Insights sends the transcripts you have already synced to eng.town on to the Claude API, where an agent reads them and writes this page. Nothing else does that. Turning it off stops the next run — insights you already have are kept, and you can delete them like anything else.
Optional. Paste an Anthropic API key and your runs are created on your account and billed to you instead of the eng.town account. It is stored encrypted and cannot be read back — not by anyone else, and not by you. To change it, paste a new one.
