Product
Tickets in. Pull requests out.
A software factory automates processes in your SDLC, not just the agent. Work can start from a ticket, an alert, a schedule, or your SDK. An agent runs it on a computer with your real stack, and you get back a pull request with the evidence attached.
How one run works
Work arrives
A ticket in Linear, an issue in GitHub, a PagerDuty alert, a cron schedule, or a webhook from your own system. The agent reads it where it already lives instead of waiting for someone to paste it into a chat box.
A computer boots with your stack on it
Your repo, dependencies, env vars, database, and services come up together. The agent starts from a working system, not an empty container.
The agent works, unattended
It reproduces the problem, changes code, runs the suite, and drives a real browser when the change needs a user to click something. You can close your laptop.
It verifies before it hands anything over
Tests run against the services that are actually up. A change that only compiles does not count as done.
You get a pull request with evidence
The diff arrives with test output, browser recordings, and the gateway log of every outbound call the agent made. Review is reading, not re-running.
What you get
Intake
Work from anywhere
Route jobs from tickets, alerts, webhooks, or schedules, or trigger a run from the API. Use the tools your team already has, or wire up your own entry point.
Execution
A computer per run
Each job gets an isolated machine that stays up for hours, so long tasks do not die when a connection drops.
Verification
Proof, not promises
Test output, screen recordings, and logs come back with the diff so you can tell finished work from a plausible-looking patch.
Control
You still own the merge
Agents open pull requests. Humans approve them. Nothing reaches your main branch without the review you already run.
Where teams use it
Common questions
What is a software factory?
A software factory automates processes in your SDLC, from intake to reviewed pull request. Work can start from a ticket, an alert, a webhook, or your own trigger. An agent completes it on a computer running your real stack, verifies the result, and returns a PR with evidence. The agent is one part; the factory is the process around it.
Do we have to change how we track work?
No. Agents read from GitHub, Linear, Jira, and Slack as they are. You do not need a separate queue or a new label scheme to get started.
How much does an agent do without a human?
It reproduces, changes code, runs tests, and opens the pull request on its own. Approval and merge stay with your team, so the blast radius is whatever your review process already allows.
What happens when the agent gets it wrong?
You get a pull request you can close. Because the run returns test output, recordings, and a gateway log, a wrong answer is cheap to diagnose instead of a mystery.
Where should we start?
Pick one workflow your team already understands well enough to grade quickly, usually nightly regression or a single PR review loop. Expand once you trust what that process returns.