This introduction explains how to use the platform step by step – from the first project idea to a supervised development run. It is deliberately hands-on and focuses on operation, not on deep architectural decisions.
The Agentic Software Factory is a local web application that lets you kick off, monitor and trace AI-assisted development processes in a structured way. In version 1 the platform uses Claude Code as the executing agent in the backend.
Instead of driving Claude Code directly from the shell, you work with a UI-centric control plane. This reduces friction and makes runs more transparent.
The platform thinks in four layers: Project, Artifacts, Team and Run.
After starting locally, e.g. via docker compose up, you reach the platform in your browser. You log in with an administrative account that is created at bootstrap via configuration.
SOFTWAREFABRIK_ADMIN_USER=admin SOFTWAREFABRIK_ADMIN_PASSWORD=ChangeMe-2026!
After login you land on the dashboard. It is the command center for projects, runs and status overviews.
Use Create new project to launch a wizard. First you provide required fields such as project title and product name. You then land directly in the editing view.
In the editing screen you maintain the actual project content: target picture, technology preferences, security requirements and free text.
This content flows directly into the generated markdown artifacts.
With a click on Generate markdown artifacts, the platform creates the specification files for the agent.
PROJECT.mdINSTRUCTIONS.mdAGENTS.mdWORKFLOW.mdDEFINITION_OF_DONE.mdREADME.mdThe platform models roles such as Architect, Developer, Reviewer, QA, Security Reviewer, Documentation and Merge/Release. In version 1 technically one executing adapter does the work, but roles and teams are already modeled at the domain level.
A run combines project, goal and optionally a team into a concrete execution. Typical fields are project, team, run title and run goal.
A new run starts out in DRAFT. You then move it to READY and start it deliberately.
While a run is in progress, you primarily work with three views:
Policies define in which phases a run may continue automatically and where manual approval is required. By default, non-critical phases are automated, while a deliberate user decision is intended before execution and completion.