01
Your machine. Your reports.
Local mode stores reports in SQLite on your machine. No server, no API key, no reporting network calls.
Open source. Closed loops.
Your agents hit friction.
Make it the last time.
Broken setup steps. Stale docs. Tools that don't work as promised.
Keep what your AI coding agents discover. Fix it at the source.
Start locally
Sign up freeFree during beta
No account. No server. Your machine.
agentfeedback / the feedback loopILLUSTRATIVE EXAMPLE REPORT
example / platformNeeds attention
The setup guide points to a command that no longer exists.
01 / What the agent found
A fresh checkout. The documented setup command. A dead end.
$ pnpm setup
ERR_PNPM_NO_SCRIPT Missing script: setup02 / A path to a fix
The current script is pnpm dev:prepare. Update docs/setup.md, then verify it on a fresh checkout.
docs / setup · suggested fix included
Evidence, context, and a next step.
Not another lost conversation.
01 / The feedback loop
A workaround is temporary.
A finding shouldn't be.
From “that was annoying” to “that is fixed.” Keep the context that usually disappears when a session ends.
- ❯ agentfeedback
kind friction
category docs / setup
✓ Report saved locally01 / Capture
Keep the evidence.
Your coding agent saves what went wrong with the docs, tools or configuration, while the context is fresh.
What goes in a report Setup command is missing
Onboarding guide is outdated
Shared cause → investigate
02 / Investigate
Find the real cause.
Review the inbox together. Group reports by cause and decide what is worth fixing.
How triage worksGuide corrected. Fix verified.
One less thing to rediscover.
03 / Resolve
Leave a clearer path.
Update the docs, repair the tool or correct the configuration. Give the next agent a clearer path.
Start locally
Illustrative workflow examples, not live report data.
02 / Boring by design
A useful inbox.
Not another platform
to babysit.
AgentFeedback fits around the way you already build. Start small. Keep control. Bring your own agent.
Read the source02
Evidence over another score.
Actual failures, concrete context, suggested fixes. No performance scores or guesses at time saved.
03
One product. Your choice.
Run locally or self-host the MIT-licensed open-source server for your team. The same client, without a new workflow.
Self-host a server03 / Start with your next session
Give your agent
somewhere to put it.
Paste this into your coding agent.
It follows the installation playbook for your harness.
Install the AgentFeedback client from https://github.com/AgentFeedback/agentfeedback and set it up for this harness.
Newsletter
Occasional product news, with an unsubscribe link in every email.
Saved until you unsubscribe or ask for removal. Privacy · Ask support to remove your address at any time.