Foundry

Foundry is a spec driven product workflow you run locally, alongside whichever coding agent you already use.

You and the coding agent write one thing: a spec, a small folder of markdown that describes your product's flows, screens, and rules. Foundry keeps that spec honest (it is machine checked, so it cannot quietly drift) and renders it into high fidelity wireframes you can click through, in the browser or as a static site you hand over.

Foundry does not call an LLM of its own. Your agent is the author. Foundry is the deterministic layer around it: the format, the checks, and the renderer.

The idea in one minute

The spec is the product. The screens are a view of it. It cannot drift.

Most workflows let the spec, the mockups, and the code wander apart until nobody trusts any of them. Foundry makes the spec the single source of truth and verifies it by machine, so the screens are always a faithful projection of the spec, never a parallel artifact that rots.

A foundry houses smiths. Three parts do the work:

• specsmith forges and verifies the spec.
• wiresmith renders the spec into screens.
• foundry orchestrates the loop around them: init, check, serve, build.

The loop

foundry init       # scaffold the spec, the design, and AGENTS.md
# author spec/ with your agent
foundry check      # verify the spec, with an exit code for CI
foundry serve      # render and click through the screens
foundry build      # the same canvas as a static site you can hand over

Four commands, and nothing else to learn. The spec is markdown in your repository, so git stays the version control and there is no second system beside it.

What Foundry is not

• Not a design tool. Beauty is a commodity; coherence and truth are the moat.
• Not a generator you pay per screen. You bring your own agent.
• Not a Figma replacement. The deliverable is a checked spec, not a canvas of rectangles.
• Not an AI wrapper. Foundry is the deterministic layer around your AI.

Open core

The local workflow, which is all of the above, is open source and free, and it is the whole of Foundry today. Your spec and screens live in your repo. You own them.

A hosted layer is planned: a private link with an access code, comments from whoever you send it to, and published versions kept side by side. That is the part teams would pay for, and it is not built yet, so nothing here pretends to do it.

Where to go next

The pages are ordered to be read in sequence, and each one stands alone if you already know where you are going.

1. Install, and what the renderer pulls in.
2. The loop, every command with its flags.
3. A brief, end to end, a real project from a vague client message to clickable screens, with the prompt used at each step. Start here if you would rather see an example than read a reference.
4. Concepts, the drift problem and what Foundry does not check.
5. Spec format and Wireframes, the two reference pages.
6. Working with an agent, wiring your agent and prompting it well.