AI coding agents can't see your dev server. They edit code, your server throws in another terminal, and the agent says "done, fixed it." Put logbear in front of your command (logbear npm run dev). Your terminal behaves exactly as before. LogBear records every run locally, and when your agent tries to finish with unread errors, it's stopped and handed the real error. Free, MIT, no account, no network calls. macOS.

Hey Product Hunt 👋

I built LogBear after one too many times an agent told me "done, everything works" while my dev server was printing a stack trace in the next terminal. The agent wasn't lying. It just couldn't see that terminal.

So LogBear sits in front of your dev command:

`logbear npm run dev`

Your terminal doesn't change: colours, Ctrl-C, breakpoints, resize all behave the same. Behind it, every run is recorded on your machine with the commit you were on. When your agent tries to finish and there are errors it hasn't read, the hook stops it and hands it the actual error. If nothing broke, it says nothing.

A few things I cared about:

• It makes no network calls. No account, no telemetry. History is SQLite files in ~/.logbear.

• It can't break your command. Output hits your screen first; kill LogBear mid-run and your app keeps going.

• psql, ssh and vim are never captured.

• `logbear diff` compares the latest run with the last one that worked: new errors, commits in between, moved lockfiles.

Works with Claude Code, Codex, pi and Hermes (blocking hook + MCP), and Cursor, Zed or anything else that speaks MCP (tools only). macOS today. Linux should build, but I haven't verified it yet.

It's free and MIT. I'd love to hear where it gets in your way, and which agent or framework you'd like it to handle better.

This hits a pain I run into constantly — the agent says everything works while the dev server died in another terminal. logbear diff is the part I like most, comparing the latest run against the last one that worked instead of just dumping logs on the agent. The local-only setup with no account or telemetry is a big plus for me too.

@jcbuilds Let me know more once you work with it for sometime and lots of improvements on this soon