Zero AI keys stored

Feedback that fixes itself.

Error trackers turn a stack trace into a fix. Nothing turns “the export button is weird on mobile” into one. fdbck triages what your users actually report and opens a draft pull request — with your own agent, on your own machine.

Free to start · no credit card · fixes are never metered

A real runfdbck#55
Reported

When i tap the Approve button up top, it 'flashes' and remains there, the Status in the Details. The In Progress in the sidenav increments +1 but the Inbox doesn't increment -1

Triaged
bugseverity 385% confidence

Approve button fails to update item status/counts on activity detail page

Shipped
…/[id]/RunLogAction.tsx+13
typecheck + lint + test1m 14s

Collect, triage, plan, ship

The work happens on your machine, with your own agent and your own credentials — fdbck never sees your code or your keys.

01

Collect

A script tag drops a feedback button into your app, or connect Zendesk, Intercom and Slack and tag what matters. Every report carries the page, the steps taken, and a PII-scrubbed snapshot.

02

Triage

Bug or feature, severity, confidence — on every report, ranked in an action-first inbox. Duplicates roll up into one item that carries the demand of all of them.

03

Plan

Work that is big or unclear gets a written approach first: what to build, which files, what still needs you to decide. You approve the plan, not the pull request.

04

Ship

Your own agent writes the fix on your machine, runs your test suite, and opens a draft PR — then tells the people who asked for it when it lands.

Everyone else stops at the ticket

Feedback tools organise a request and hand it to you. Coding agents write code from an issue you already wrote. Error trackers only see what crashed. fdbck is the part in between.

ToolStarts fromEnds at
Canny, ProductboardA typed requestA prioritised ticket
Sentry SeerAn error eventA patch
LogRocketA session replayReproduction steps
Claude Code, CodexAn issue you wroteA diff
fdbckA typed requestA reviewed draft PR

Every product above is good at what it was built for. The row that matters is the one nobody else spans.

Built to be trusted unattended

Nothing ships without you

91%

of fix attempts finish and produce a pull request. Runs paused by your agent's own usage limits retry rather than counting as failures.

Zero

model keys stored. The work runs on your machine with your subscription, so fixes and PRs are never metered or charged per use.

Draft

Every PR opens as a draft against an isolated branch, labelled and linked back to the report. Auto-merge is off until you turn it on.

Your code, your machine

fdbck stores no model keys, and never sees your source.

A small local worker does all the model and git work on your machine, with your own Claude, Codex, or Gemini and your own GitHub credentials. fdbck holds the feedback and the state of your queue — not your code, your history, or your logins.

Your inbox is waiting to clear itself.