Welcome to the ZeroWidth docs
ZeroWidth helps you find where AI changes your business, build the software to ship it, and measure whether it works. We design the solutions and ship the tools that make them real. These docs cover both — the tools you can plug into directly, and the integrations that connect them to the rest of your stack.
Where to start
Design AI behaviors as flows, then run them from your code. The flow is the contract — the model and prompt underneath can change without breaking your integration.
ProductCaliperEvaluation for AI features that have to keep working. Datasets, rubrics, human reviews, and LLM-judge evals you can wire into CI.
ProductCompassMap what your team actually does as a living knowledge graph, and surface the opportunities worth trying.
ProductLedgerDecision memory. Record calls with pre-registered expectations, settle them against evidence, and keep the lessons findable.
ProductNapkinThe shared sketchbook — draw together in real time, drop in live workspace objects, and let zv1 look at the board.
ReferencePublic APIAuthentication, key management, scopes, cost caps, and the shared error envelope — the setup every API caller needs.
IntegrationMCP serverPlug these docs — and your Compass map — into Claude Desktop, Cursor, or any LLM client that speaks Model Context Protocol.
Reading paths
Different readers want different things. Three common shapes:
- Evaluating ZeroWidth for the first time
Skim the product overviews — Workbench, Caliper, Compass, Ledger, and Napkin — or the full catalog. Then read the Declaration of Principles for how we think about AI work, and the Distributed Cognition notebook for the longer-form writing.
- Integrating ZeroWidth into your stack
Make your first API call in a couple of minutes, then go deep on the API reference — auth, errors, pagination, idempotency. Call a flow from your code with the Workbench run endpoint, and set up the MCP server so your LLM tooling can reach ZeroWidth in-context.
- Wiring evaluation into CI
Read the Caliper API reference — fetch an eval, submit your model's actuals, and read the score back. Built for the CI-scoring loop.
What's NOT here
A few things this docs surface is not trying to be:
- A blog or newsletter. Long-form writing lives at Distributed Cognition.
- A marketing pitch. That register lives on the main zerowidth.ai site. These docs are reference + how-to.
- A status page. Live service status is at zerowidth.statuspage.io.
Need help?
We're small and direct. Email contact@zerowidth.ai for anything — feature requests, integration questions, "is this a bug or am I holding it wrong" — and a human gets back within a business day.