Platform
You say it. We build it.
ForgeOS picks the AI and tools, does the work, checks it, and shows you.
Six steps. Tap one.
01Understand
Your words become a short list you can fix.
02Choose
Each step gets the right AI. Light first, stronger if needed.
03Do
Work happens in a copy. Your project isn’t touched.
04Check
Your tests run. Pages open on a phone and a laptop.
05Prove
You see the exact change, and anything that failed.
06Deliver
You accept, ask for changes, or publish. You can undo. Projects it made can accept themselves when tests pass — you can turn that off. Publishing always waits for you.
37 kinds of project, from one sentence.
Show the list
Software
Available here
- API services
- Python tools
- Node services
- Go services
- Ruby tools
Experimental
- Web apps
- AI apps
- Progressive web apps
- Data pipelines
- Serverless functions
- Browser extensions
- Scrapers
- Shell automation
- 3D scenes
- 2D games
- 3D games
- Hardware apps
Sites
Available here
- Websites
- Landing pages
- Blogs
- Portfolios
- Docs sites
Experimental
- Shops
- Campaign sites
- Government services
Design
Experimental
- Brand assets
- Social kits
- Print pieces
- Motion
- Prototypes
- Live performances
Documents & data
Available here
- Dashboards
Experimental
- Slide decks
- Data reports
- Newsletters
- Press kits
- Impact reports
All can start. 37 have test briefs. A model built 11 to a checked result. What the counts mean → See every kind as a card, with an ask that starts it →
Works with what you use.
Every connection starts read-only.
GitHub Experimental
Read your repos. Open a pull request after you accept.
MCP servers Experimental
Connect any server with sign-in.
Google Workspace Experimental
Gmail, Drive and Calendar, read-only.
Your AI keys Experimental
OpenRouter, Anthropic, OpenAI, Gemini, xAI, Mistral, DeepSeek or a local model.
Slack & more Not available yet
Coming one by one.
Available here: the connection is proven live in our capability record. Experimental: it is built and checked, but no record proves it live end to end. Not available yet: nothing here connects it. How each label is decided.
Nothing hidden. Export it all.
What we keep for each goal
# Example goal "Fix the failing checkout test" model chosen per step · recorded copy isolated · your project untouched tests 14 passed · 0 failed change 2 files · +38 −9 decision accepted by you → pull request