Software Factory
Software Factory is where you follow a piece of work from a plain request all the way to shipped code. A card gets worked by an agent on the Development board. The agent's output becomes a Pull Request you review. That pull request (or another external trigger) produces a Deployment you can track. And at any point you can drop into the agent's actual Local Environment to read the code yourself.
Think of it as the core idea of the platform, made concrete page by page: the AI does the labor inside a sandbox, and you approve the decisions that matter. You stay on the loop, not in it.
In this section
- Development: the Kanban board, creating and opening issues, subtasks, Card review and Human input checkpoints, Stop and Reject, and finishing work.
- Pull Requests: the list and its filters and rules, and reviewing a pull request (Files/Analysis/History, Approve/Reject/Merge).
- Deployments: the list and its filters and rules, and the deployment detail page (Analysis/Details/Jobs/Artifacts).
- Local Environment: opening the agent's own cloud workspace for a card.
The workflow, in order
- Capture the work. On the Development board, a card starts in Backlog, created by you through New Issue, or opened automatically by an agent from an incident, a pull request, or a deployment.
- Queue it, then dispatch it. Moving a card to Todo and then into In Progress (by drag, or the card's Start banner) is what actually kicks off an agent run.
- The agent works, and may stop to ask. While In Progress, an agent plans and implements inside a private sandbox. It can raise a Card review (approval, questions, confirmation) or a Human input request at any point. Both are answered from the card's own detail page.
- Review lands in Review. Every checkpoint response either sends the agent back to work, or moves the card on. You can also Stop a live run or Reject a card outright at any time.
- The card reaches Done, and code becomes a pull request. If the work was built in a scratch sandbox, Save & ship pushes it to a real repository; either way, the resulting change shows up on Pull Requests, linked back to the card.
- You review and decide on the pull request. Run its AI analysis, read the diff, and Approve, Reject, or Merge, all from the pull request's own review page.
- Merging (or any other trigger) produces a deployment. Track it on Deployments, where you can run failure analysis and, if it is not already linked, turn the run into a new card.
- Anywhere along the way, drop into the real code. Local Environment opens the exact cloud workspace an agent used for a card, in VS Code, not a copy of it.
Where to find it
Left sidebar → Software Factory, containing Development, Pull Requests, Deployments, and Local Environment (plus Planning and Testing, both marked "Soon" and not yet functional). Every page in this section is scoped to whichever project you currently have selected in the sidebar.