Retrace

Your real users are your QA team. Retrace finds the bugs they hit, writes the tests, and opens the fix PR.

What it does

• Capture live user sessions from the browser SDK or PostHog replay.
• Detect UX failures: console errors, broken network calls, rage clicks, dead clicks, blank renders, error toasts, and session abandonment.
• Cluster repeated failures into replay-backed issues with severity and affected users.
• Auto-generate deterministic UI regression specs so real bugs become repeatable tests.
• Fix by linking each issue to a repo, ranking likely source files, and opening a draft PR with a fix prompt.
• Track issues as new, ongoing, regressed, or resolved, and re-run specs after fixes.

Install

curl -fsSL https://raw.githubusercontent.com/txmed82/retrace/master/install.sh | bash

Or with Docker Compose:

curl -fsSL https://raw.githubusercontent.com/txmed82/retrace/master/install.sh | bash -s -- --docker

Then:

retrace quickstart   # mints an SDK key and prints a ready-to-paste <script> tag
retrace api serve     # start the ingest API
retrace ui             # open the local dashboard

Quick demo

retrace github connect --repo your-org/your-app --local-path /path/to/repo
retrace qa auto --repo your-org/your-app

This picks the highest-priority open incident, reproduces it with a generated test, scores the repo, and opens a draft fix PR.

Why Retrace

Problem	How Retrace helps
Users find bugs first	Session replay + automatic detection
Reproducing bugs is slow	Replay-derived regression specs
Debugging takes hours	Fix prompts with replay evidence + likely source files
Too many tools	One incident queue for replay / UI test / API test / error monitor / PR review
Budget constraints	Self-hostable, BYOK for LLM and third-party services

Architecture

Browser / PostHog → Ingest API → Replay Processing → Detectors → Clusterer →
  Unified Incidents → Auto-Repro → Repo Scoring → Fix Prompt → Draft PR

Documentation

• Install
• Usage
• Quickstart
• Architecture

License

MIT