How ZeroWidth fits together

ZeroWidth is a suite of tools around one idea: find what's worth trying, build it, prove it works, and remember what you learned — then do it again. This page is the map of how the pieces relate and what the words mean, so the rest of the docs read faster.

Everything lives in a workspace

A workspace is the boundary around your work — its flows, datasets, pages, members, keys, and billing. Work in one workspace can't reach into another. You'll usually have one per team or product; each member has a role (owner, admin, member) that sets what they can do.

The suite

Each tool earns its keep alone; together they're a loop.

• Compass — find. Map how your team actually works — the workflows, the people, the systems, the friction — and surface the opportunities: where an AI build, or a simpler change, could improve the work.
• Workbench — build. Design an AI behavior as a flow and call it from your code. The flow's behavior is the contract; the model underneath can change without breaking your integration.
• Caliper — prove. Score a flow (or any model) against a rubric on a dataset you trust, run after run, so a regression shows up as a falling number instead of a support ticket.
• Ledger — remember. Record decisions with pre-registered expectations, settle them against evidence, and keep the lessons where the next decision can find them.
• Napkin — sketch. A shared whiteboard where rough thinking happens — with your team live on the board and zv1 able to look at what you drew.
• Prism — explore, then ask. Map the space around an idea, then take it to real people — studies, AI-led interviews, panels — and read the evidence that comes back.

And in every tool, zv1 — the assistant that works across all of them.

What Caliper measures points Compass at the next thing to build, and Ledger records whether the call was right. That's the improvement loop — the reason the suite is one product. The full catalog is the canonical list.

Two shared surfaces

Whatever you build reaches your code two ways:

• The public API — plain HTTPS + JSON at api.zerowidth.ai/1.0. Run flows, submit eval runs, browse public flows.
• The MCP server — the same capabilities as tools for Claude, Cursor, or any MCP client, so an LLM can drive your workspace directly.

Both authenticate with a bearer token — an API key for services, a personal access token for you.

The vocabulary

The nouns you'll meet across the docs:

Where to start