Introduction
Cat-Factory is a self-hosted platform for designing software architecturally and having LLM agents build it autonomously. You lay out your system as a visual board of services, modules, and tasks, then run agent pipelines that produce real, reviewed pull requests, with full observability over every step.
Who it's for
Cat-Factory is built for teams that want to scale software delivery with agents while keeping humans in control of cost and quality. It fits well when you want to:
- Turn an architecture sketch directly into implementation work, without maintaining a separate backlog.
- Have agents do the repetitive build-out while engineers focus on design and review.
- Keep a tight feedback loop: every change lands as a pull request you approve before merge.
- Run everything on your own infrastructure (Cloudflare or Node.js), with your own model providers.
What makes it different
Many "AI coding" tools stop at generating snippets, and even the autonomous ones treat a flat task list as the plan. Cat-Factory is built around two ideas that change what "done" means:
The board is the plan. Services, modules, and tasks form a hierarchy (frame -> subframes -> leaves) with dependency edges. The same board is your design artifact and your unit of work.
Agents do real work through pull requests. Implementation phases run actual coding agents on repository checkouts. Completion is defined by merged PRs with passing CI, not by generated text.
What you get
| Capability | What it means for you |
|---|---|
| Visual planning | A pannable, zoomable canvas with frames, modules, and tasks; drag-and-drop reparenting and dependency edges. |
| Agent pipelines | Reusable, ordered chains of agent steps with per-block model selection and human decision points. |
| Real code | Agents clone repos, implement work, and open PRs; merges flip blocks to "done". |
| Requirements | A reviewer agent finds gaps and risks per task; a requirements-writer keeps a unified, in-repo spec (with Gherkin scenarios) for the whole service. |
| Shared services | A service is account-owned and can be mounted onto many teams' boards in an org as one shared, synced copy. |
| Recurring pipelines | Schedule maintenance (dependency updates, tech-debt passes) to re-run a pipeline on a cadence. |
| Integrations | GitHub App for repo/PR/issue operations; import context from Jira, GitHub Issues, Confluence, Notion, and GitHub repo docs. |
| Spend control | Organization-wide monthly LLM budget with metering, prompt caching, auto-pause at the cap, and rollover resumption. |
| Observability | WebSocket event streaming of every step, decision, failure, and spend update, with no polling. |
How it fits together
At a glance, Cat-Factory is:
- A Nuxt single-page app that renders the board on a Vue Flow canvas.
- A runtime-neutral backend (Hono HTTP layer) that runs on either Cloudflare Workers (with D1, Durable Objects, and Workflows) or Node.js (with PostgreSQL and pg-boss).
- Per-run containers that execute the actual coding work and Git operations.
See Core Concepts for the vocabulary, or jump straight to the Quick Start to get an instance running.