softwarefabrik.io
Product · v0.52.7

What the factory is — and what it does for you.

The Agentic Software Factory is a web application on your own machine. You describe what should be built. AI agents such as Claude Code or Codex build it, the factory checks the result — and you approve. Every step stays traceable.

See the demo → Install Pricing

The problem

Coding agents write in minutes code that takes a human days. Used directly in the shell, though, the process around them is left to chance:

  • Tasks without structureProject descriptions are written on the side in a chat, are not versioned and are lost by the next project.
  • Large, unclear changesLong sessions produce changes nobody can review sensibly any more.
  • Approvals without contextWhoever approves often cannot see what changed or whether build and tests are green.
  • No traceabilityLogs, results and decisions are scattered across terminal, editor and notes.

The factory takes over exactly this part: it gives every run a frame, checks the result and records who approved what. The creative work stays with the agent.

Who it is for

Individual

Developers and architects

With your own Claude or Codex subscription on your own machine — more structured than the shell, no server needed.

Team

Tech leads and small product teams

Runs, logs and approvals in one place; follow-up runs instead of overlong single sessions.

Organisation

Companies and public bodies

Tenants, roles, SSO, a signed audit chain and policy-as-code; self-hosted or air-gap via Enterprise.

If you only send a prompt now and then, Claude Code on its own is faster. The factory pays off as soon as results have to be checked, repeated or shown to someone.

What it looks like

localhost:8080/wizard/schnellstart
The quickstart form: name, kind and description, then “Bauen” (German UI)
Describe: in the quickstart, a name, a kind and a description in your own words are enough. (German UI)
localhost:8080/runs/…
The run page: status, progress, “Was jetzt?” and approval with one button (German UI)
Build and approve: the run page shows progress, “Was jetzt?” (what now?) and the one button you need right now.
localhost:8080
The Software Factory dashboard with metrics, license status and notices (German UI)
Overview: the dashboard with projects, runs, costs and license status.

The loop: describe, build, check, approve, keep improving

  1. DescribeYou say in your own words what should be built or changed — for a new project or an existing one from a folder or Git address. If you want to choose stack and checking tools yourself, use the detailed wizard.
  2. Build“Bauen” (build) starts a run. The AI agent works in the background in your project folder, on its own branch; the factory makes the commits itself.
  3. CheckThe factory builds and tests the result itself. If the build is red, the agent corrects it a limited number of times. Then the quality check rates the result with several reviewers — advisory or blocking.
  4. ApproveDepending on the guardrails you confirm before execution or before the merge — looking at the changes, with one button. With a Git remote and a GitHub token a pull request is opened, otherwise the factory merges locally.
  5. Keep improving“Weiter verbessern” (keep improving) leads to the next task, “Vorschläge machen lassen” (get suggestions) puts ideas into the backlog. A project grows in small, reviewable steps.

What works today

As of v0.52.7 — delivered and visible in the demo:

  • Creating projects: quickstart with three inputs, detailed wizard with 19 templates, import of existing projects.
  • AI agents: Claude Code, Codex, Gemini, Kimi and Aider via subscription or API key, plus a local model and a free mock for trying things out.
  • Runs: in the background, with approval, live logs, costs per run and the result as a ZIP.
  • Checking: build check per project profile, quality check with reviewers for architecture, security and dependencies.
  • Evolving: plan runs, backlog, a branch per run, pull requests on GitHub, project memory.
  • Team operation: tenants, roles, SSO, signed audit chain, policy-as-code and audit export.

What does not work yet or only with restrictions — such as parallel runs (feature flag only) or the experimental cloud gateways — is listed openly under Known limitations.

Pricing

Three tiers on the same software: Community is free, Professional and Enterprise extend limits, support and governance. You pay for the AI models directly to the provider. Prices, limits and lease durations are on the pricing page.

Try it yourself

Click through the demo without installing anything — or request the single-seat starter package and start your first run in the quickstart.

See the demo → Request starter package

Read on: Architecture · Glossary and help · Whitepaper · News