The software factory, self-hosted.

Issues, todos and suggestions become planned changes, run in isolated build cells, and land in GitHub as pull requests. ProPR runs that whole line as free, Apache 2.0 software on your server. You decide whether green pull requests wait for a person or merge on their own.

What counts as a factory

  • Planning that turns issues, todos and suggestions into scoped work.
  • Parallel, isolated runs.
  • Pull requests with execution records.
  • Review with scored findings and CI.
  • Refinement from comments and failed checks.
  • Merge by policy, with telemetry across the whole line.

Overview

Six stations, one loop.

Plan, isolated run, pull request, review, refine, merge. Every ProPR change follows the same line.

Six-station delivery line Merge policy visible
Telemetry across the linePrompts, logs, model usage, cost, duration, commits, and recovery stay attached to each task.
Work enters as a plan on the left. Each run reaches GitHub as a pull request before review and refinement begin, then the selected repository policy controls the merge.
A generated plan broken into five ordered tasks, each a future GitHub issue

1 · Plan

GitHub issues, repository todos, improvement suggestions and PR comments all become work. Planning turns each into review-sized tasks with the context visible, and your approval is the handoff.

The end of an isolated run: the restricted network denied three hosts, the agent finished, and ProPR created the pull request

3 · Pull request

Each run opens a normal GitHub pull request with its summary and execution record. Review, CI and follow-up work happen on that branch.

One pull request’s runs: review scored 5, a fix, review scored 7, and the Ultrafix cycle in progress

4 · Review

/review posts scored findings, and Ultrafix runs review and fix cycles toward your target score, within your cycle limit. CI runs as on any branch.

A review scored 6 with two issues, then the /fix run that applies finding F1 and suggestion S2

5 · Refine

Plain PR comments become follow-up commits on the same branch through PR fine-tuning. /fix applies review findings, and a failing check starts its own fix.

Execution settings for every task in the plan: agent and model, Ultrafix with a minimum review score of 8/10 and a loop limit, and auto-merge

6 · Merge

Merge in GitHub, or let green pull requests auto-merge. Larger changes run as an Epic-mode chain, each child PR merging before the next task starts.

Telemetry across the line

Every task keeps its prompt, diff, logs, model, tokens, cost and duration on the dashboard and CLI, so you see where the line is slow, expensive or wrong.

LLM log rows tied to their work: two pull requests, a plan and a repository summary, each with its tokens, cost and the capacity its call used
The LLM log ties each call’s tokens and cost to the task, plan, or repository that spent them.

You own the factory.

The whole line is Apache 2.0 software on a Docker host you control, running Claude Code, Codex, Antigravity CLI, Mistral Vibe or OpenCode on your subscriptions or API keys with no token markup. Compare that with hosted agent-workforce platforms.

Repository policy closes the loop.

Comments refine each branch until it is ready. Then you merge by hand, or auto-merge completes green pull requests, including the child PRs in an Epic chain.

A one-person factory, running in public since May 2025.

ProPR builds itself on this line: 2,400+ pull requests opened, 2,100+ merged, 690+ of them building ProPR itself. Every review score and merge is public.