AI agents that turn your tickets into pull requests, with you in control
PRFactory takes a ticket through planning, tests, code and AI code review to a pull request. A team of your own AI coding agents does the work on your machine. You answer the questions in the browser, and nothing merges until a person says so.
beat 1s
idle since 2:15
beat 4s
An epic, cut into slices you can actually start
Wave-ordered. Every acceptance criterion owned exactly once. One decision round instead of dozens of interruptions.
D-7 Should a child agent be allowed to spawn children of its own?
How It Works
1.Bring your agent
Connect Claude Code, Codex or any MCP client, and your Jira, GitHub, Bitbucket or Azure DevOps. Your agents run on your machine, on the subscription you already pay for.
2.Plan the epic
An epic is cut into wave-ordered slices, every acceptance criterion owned exactly once, and the open questions are collected into one decision round.
3.Watch the team build it
A lead agent spawns the team, and the ticket shows every agent's role, state and output live. Message any one of them, or stop it.
4.Approve at every checkpoint
Answers, plan, test plan, review and the pull request are all yours to approve. The team waits; it never merges on its own.
Ticket to reviewed pull request, in the open
The run above ends here. Every phase leaves an artefact you can read, and a checkpoint you can refuse.
The seven-phase flow is the built-in process. Define your own beside it.
A process is a list of named steps. Each step picks an agent backend, a model and a reasoning effort, and names the file it has to produce. Named outcomes choose the next step, so the process branches instead of running straight through. Mark a step read-only to keep it away from your code, put a human checkpoint wherever approval matters, and bind the repositories it works in. A support-triage starter ships in the box.
Step 0: Classify the case
read-onlyYou are a support triage analyst. Work read-only: never commit, branch, push, or open a PR.
Ticket {{TicketKey}}: {{Title}}
Write your full analysis to classify.md. End classify.md with a fenced outcome block naming exactly one of: user-error, config-error, bug.
Key Features
Live agent teams
A lead agent runs a team of child agents in your own harnesses. Their roles, states and output are on the ticket while they work, and you can message or stop any one of them.
Epic planning that holds
An epic is cut into wave-ordered, build-sized slices with every acceptance criterion owned exactly once. Coverage is proven from the map, not asserted in a summary.
Bring your own agent
Claude Code, Codex or any MCP client, via the built-in MCP server or the local worker. They run on your machine, on the subscription you already have: no credits, no API keys to hand over.
Human checkpoints that hold
Answers, plan, test plan, review and PR are approvals, not notifications. The team parks and waits; agents never merge on their own.
Your own workflows
Define processes beyond the seven-phase default as portable JSON graphs: an agent, a model and an effort per step, outcomes that pick the next step, and human checkpoints where you want them. A support-triage starter ships in the box.
A fleet of workers
Every developer's machine joins a fleet you can see. A ticket stays on the machine that started it, hands over with the work preserved, and dependencies decide what may be built next.
Multi-platform
GitHub, Bitbucket, Azure DevOps and Jira, across one or several repositories per work item. Integrate with the toolchain you already run.
Enterprise security
Multi-tenant isolation and encrypted credentials. Your agents run on your machine; PRFactory keeps the workflow artefacts.
From one developer's agents to a whole team's machines
How it works
Seven phases from ticket to pull request, which agent runs each one, and every point where a person approves.
AI agent teams on your machine
The worker CLI runs a lead agent and its team locally. The server orchestrates and never runs an agent.
Product owner and developer, one ticket
Clarifying questions are answered in the browser while the coding happens on another machine.
Work spread across machines
Pin a ticket to a machine, or move it with its branch pushed, so workload and token spend are shared.
Questions people ask about PRFactory
What is PRFactory?
Does PRFactory run the AI agents itself?
Which AI coding agents does it work with?
Do I need an API key or LLM credits?
Does PRFactory do AI code review?
Which Git platforms are supported?
Where is PRFactory hosted?
Put a team of agents on your next epic
Bring the agents you already use. Keep every decision.
Get started free