# ZeroWidth

> ZeroWidth finds where AI changes your business, builds the software to ship it, and measures whether it works.

Full documentation at https://zerowidth.ai/docs. Any page can be copied as Markdown from its "Copy page" menu.

## Software

- [Chat](https://zerowidth.ai/software/chat): Work it out with your teammates and your agents, in one place.
- [Compass](https://zerowidth.ai/software/compass): Map how your business actually runs.
- [Prism](https://zerowidth.ai/software/prism): Explore an idea, then find out what people really think.
- [Workbench](https://zerowidth.ai/software/workbench): Build the behavior: software your team can read and change.
- [Caliper](https://zerowidth.ai/software/caliper): Turn “does it work?” into a hard number you can ship against.
- [Ledger](https://zerowidth.ai/software/ledger): Remember what worked, what didn't, and why.
- [Napkin](https://zerowidth.ai/software/napkin): Sketch it out with your team and get everyone looking at the same thing.

## Company

- [Pricing](https://zerowidth.ai/pricing): The plans, what each includes, and how inference is billed.
- [Security and trust](https://zerowidth.ai/trust): Where your data is stored, what models see, and the commitments we make in writing.
- [Integrations](https://zerowidth.ai/integrations): The tools and models ZeroWidth connects to, and the ways in for your own agents and code.
- [About](https://zerowidth.ai/about): Who we are and the principles we hold ourselves to.

## Overview

- [Introduction](https://zerowidth.ai/docs/overview): What ZeroWidth ships, who these docs are for, and where to start.
- [Make your first API call](https://zerowidth.ai/docs/overview/quickstart): From zero to a live response against the ZeroWidth API in a couple of minutes.
- [Built on Workbench](https://zerowidth.ai/docs/overview/built-on-workbench): Every intelligent thing ZeroWidth does runs on a Workbench flow — the same engine you build on.
- [How ZeroWidth fits together](https://zerowidth.ai/docs/overview/concepts): How ZeroWidth fits together — workspaces, the suite, the shared surfaces, and the vocabulary.
- [Part 1 — Build the flow](https://zerowidth.ai/docs/overview/tutorial/build-a-flow): Build an AI classifier that sorts a support ticket into the right queue.
- [Part 2 — Call it from your code](https://zerowidth.ai/docs/overview/tutorial/call-it-from-code): Call the support-ticket classifier from code with an API key and read back the queue, so your ticketing system can route tickets on its own.
- [Part 3 — Prove it holds up](https://zerowidth.ai/docs/overview/tutorial/evaluate-it): Score the classifier against real tickets so a regression can't quietly ship.
- [Capstone — The whole loop](https://zerowidth.ai/docs/overview/tutorial/the-whole-loop): Every tool in the suite, on one problem — from a napkin sketch to a settled lesson.
- [The improvement loop](https://zerowidth.ai/docs/overview/guides/the-improvement-loop): How Compass, Workbench, and Caliper compose into one loop — find the change worth making, build it, prove it works, repeat.
- [Handle API errors well](https://zerowidth.ai/docs/overview/guides/handling-errors): A small, robust error-handling strategy that covers everything the API throws at you.
- [Troubleshooting](https://zerowidth.ai/docs/overview/troubleshooting): Symptom → cause → fix for every error the API returns — auth, quota, rate limits, timeouts, and streaming.
- [Example projects](https://zerowidth.ai/docs/overview/guides/example-projects): Working starting points — call flows from Node, Python, and the browser; batch data through them; run them in your own infrastructure.
- [Public API](https://zerowidth.ai/docs/overview/api): Authenticate, manage keys, and call ZeroWidth services from your own code.
- [Authentication](https://zerowidth.ai/docs/overview/api/authentication): Bearer-token authentication, key format, and where keys are issued.
- [API keys](https://zerowidth.ai/docs/overview/api/keys): Mint, scope, budget, and revoke API keys from the workspace dashboard.
- [Personal access tokens](https://zerowidth.ai/docs/overview/api/personal-tokens): User-identity tokens for the CLI and MCP — they act as you, across your workspaces.
- [Errors](https://zerowidth.ai/docs/overview/api/errors): The shared error envelope, the closed code vocabulary, and how to handle them.
- [Limits](https://zerowidth.ai/docs/overview/api/limits): Per-key cost caps, workspace inference spend, request rate limits, and run-duration ceilings.
- [Pagination](https://zerowidth.ai/docs/overview/api/pagination): How list endpoints page — the cursor envelope every collection follows.
- [Idempotency](https://zerowidth.ai/docs/overview/api/idempotency): Send an Idempotency-Key header so a retried write replays the original response instead of running a flow twice or charging twice.
- [Streaming](https://zerowidth.ai/docs/overview/api/streaming): Stream a flow run as server-sent events instead of waiting for one JSON blob.
- [Versioning](https://zerowidth.ai/docs/overview/api/versioning): How the API is versioned, what changes are safe, and how deprecation works.
- [Workspaces](https://zerowidth.ai/docs/overview/account/workspaces): The container for everything you build — and who can do what inside it.
- [Your workspace](https://zerowidth.ai/docs/overview/account/files): The front door to a workspace — your files, your folders, and what's waiting for you.
- [Invite your team](https://zerowidth.ai/docs/overview/account/team): Invite people into a workspace by email or a link you share, and choose the role that sets what they can do.
- [Plans & billing](https://zerowidth.ai/docs/overview/account/plans-and-billing): How plans work: seats, included inference, knowledge base storage, API rate limits and support across the Free, Starter, Pro and Enterprise tiers.
- [Spend & inference credit](https://zerowidth.ai/docs/overview/account/spend-and-credits): What draws down your included inference and prepaid balance, and what happens when it runs out.
- [MCP server](https://zerowidth.ai/docs/overview/mcp): Drive ZeroWidth — and search its docs — from any LLM client that speaks Model Context Protocol.
- [CLI](https://zerowidth.ai/docs/overview/cli): Your workspace in the terminal: list and run flows, chat with one, search the docs and call tools from a shell, a script or a CI job.

## Zero

- [Zero](https://zerowidth.ai/docs/zv1): Zero is the assistant that helps you build AI with ZeroWidth: what's worth changing, how to wire up a flow, and what an eval is telling you.
- [Using Zero](https://zerowidth.ai/docs/zv1/using-zv1): Where to find the assistant, what to ask it, and how it keeps context.
- [Approvals](https://zerowidth.ai/docs/zv1/approvals): Zero shows you what it's about to change and waits for your OK.
- [Inbox](https://zerowidth.ai/docs/zv1/inbox): Where Zero puts anything it can't finish without you, with a count on the tab whenever something is waiting.
- [Rooms](https://zerowidth.ai/docs/zv1/rooms): Where a group works on something together, with Zero and your flows in the conversation.
- [Sidebars](https://zerowidth.ai/docs/zv1/sidebars): Take one question aside, settle it, and bring the answer back to the room.
- [Routines](https://zerowidth.ai/docs/zv1/routines): Things you'd otherwise remember to ask Zero — running on a schedule instead.
- [Skills](https://zerowidth.ai/docs/zv1/skills): How Zero approaches each tool, and how it learns the way you work.

## Caliper

- [Caliper](https://zerowidth.ai/docs/caliper): Turn \"does it work?\" into a number: collect the outputs that matter, score them against a rubric your team agrees on, and watch the baseline over time.
- [Quickstart](https://zerowidth.ai/docs/caliper/quickstart): Get a dataset in, score it against a rubric, and read the result — even if you're starting from nothing.
- [Datasets](https://zerowidth.ai/docs/caliper/datasets): Collections of items scored in a review, eval, or spec — and the slot model that decides which.
- [Rubrics](https://zerowidth.ai/docs/caliper/rubrics): The criteria raters score each item against, each with a name, a description and a numeric scale. Start from a preset rubric or write your own.
- [Reviews](https://zerowidth.ai/docs/caliper/reviews): Pair a dataset with a rubric, collect ratings from humans, read the results.
- [Evals](https://zerowidth.ai/docs/caliper/evals): Automatic scoring — an LLM judge grades a dataset against a rubric, run after run.
- [Specs](https://zerowidth.ai/docs/caliper/specs): The inverse of an eval — gather what the ideal answer should be, then score against it.
- [Sources](https://zerowidth.ai/docs/caliper/sources): See what your agent does on real traffic: every conversation an app sends Caliper, with each model call, tool call and lookup in between.
- [Themes](https://zerowidth.ai/docs/caliper/analysis): Read across every conversation in a dataset and find out what people actually use it for.
- [Sharing & invites](https://zerowidth.ai/docs/caliper/sharing): Bringing raters into a review — teammates by email, anyone by link.
- [Run evals in CI](https://zerowidth.ai/docs/caliper/guides/run-evals-in-ci): Score a model that runs in your own infrastructure on every pull request, and fail the build when the score drops.
- [Set up a human review](https://zerowidth.ai/docs/caliper/guides/human-review): Get structured ratings from the people whose judgment counts — and read where they agree and don't.
- [Compare runs & read heatmaps](https://zerowidth.ai/docs/caliper/guides/compare-runs): Tell a real regression from noise, and find exactly what moved.
- [Generate a dataset with AI](https://zerowidth.ai/docs/caliper/guides/generate-dataset): Draft a test set from a brief — including answers and adversarial cases — then refine.
- [Send traces with OpenTelemetry](https://zerowidth.ai/docs/caliper/guides/send-traces-with-opentelemetry): Send OpenTelemetry spans from the agent framework you already run to Caliper by changing configuration. Each trace becomes one conversation.
- [Troubleshooting](https://zerowidth.ai/docs/caliper/troubleshooting): Diagnose eval runs — failed runs vs failed items, judge errors, CI submissions, and share links.
- [Caliper API reference](https://zerowidth.ai/docs/caliper/api): Submit external evaluation runs and score them against a rubric — programmatically.
- [GET /1.0/caliper/evals/:id](https://zerowidth.ai/docs/caliper/api/get-eval): Fetch an eval and its bound dataset items.
- [POST /1.0/caliper/evals/:id/runs](https://zerowidth.ai/docs/caliper/api/submit-run): Submit pre-computed actuals for Caliper to score.
- [GET /1.0/caliper/evals/:id/runs/:runId](https://zerowidth.ai/docs/caliper/api/get-run): Poll one run until it settles and read its scores.
- [POST /1.0/caliper/feedback](https://zerowidth.ai/docs/caliper/api/submit-feedback): Log what a person thought of one AI response, from anywhere.
- [POST /1.0/caliper/traces](https://zerowidth.ai/docs/caliper/api/send-traces): Send what your agent did, from anywhere.

## Workbench

- [Workbench](https://zerowidth.ai/docs/workbench): Design AI behaviors as flows — then call them from your code.
- [Quickstart](https://zerowidth.ai/docs/workbench/quickstart): From zero to a flow you can call from your code, in about five minutes.
- [Flows](https://zerowidth.ai/docs/workbench/flows): What a flow is, how you start one, and what you pin in production.
- [Input & output](https://zerowidth.ai/docs/workbench/input-output): How a flow's input and output nodes work: each carries a key your code uses to supply a value or read a result.
- [Nodes](https://zerowidth.ai/docs/workbench/nodes): The building blocks you compose on the canvas — models, agents, logic, and more.
- [Knowledge bases](https://zerowidth.ai/docs/workbench/knowledge-bases): Give your flows something to know — searchable documents, queryable tables, and walkable knowledge graphs.
- [Integrations](https://zerowidth.ai/docs/workbench/integrations): How a flow reaches the web and the systems your team already runs.
- [Imports](https://zerowidth.ai/docs/workbench/imports): Reuse one flow inside another as a pinned, versioned building block.
- [Templates](https://zerowidth.ai/docs/workbench/templates): Publish a flow for anyone to read and fork — or start from someone else's.
- [Schedules](https://zerowidth.ai/docs/workbench/schedules): Run a published flow on a cadence — hourly, daily, or weekly — with results delivered where you'll see them.
- [Tasks](https://zerowidth.ai/docs/workbench/tasks): Write a process as one sentence. Flows do the work, people take the decisions, and every run reads at a glance.
- [Build an agent](https://zerowidth.ai/docs/workbench/guides/build-an-agent): Build a chat assistant that can use tools — then call it from your code.
- [Build a structured extractor](https://zerowidth.ai/docs/workbench/guides/build-a-structured-extractor): Turn messy text into a clean category or a typed JSON object — reliably, every time.
- [Reuse logic with subflows](https://zerowidth.ai/docs/workbench/guides/reuse-with-subflows): Pull a chunk of a working flow out into its own reusable flow — in one step.
- [Ship a flow to production](https://zerowidth.ai/docs/workbench/guides/ship-to-production): Pin a version so your integration doesn't change every time someone saves the flow.
- [Stream a live response](https://zerowidth.ai/docs/workbench/guides/stream-responses): Show tokens as they generate instead of waiting for the whole run — a live chat feel.
- [Evaluate a flow with Caliper](https://zerowidth.ai/docs/workbench/guides/evaluate-with-caliper): Go from a working flow to a measured score — and keep an eye on it from Workbench.
- [Inspect a run](https://zerowidth.ai/docs/workbench/guides/inspect-runs): See exactly what a flow did — node by node, inputs and outputs — in the flow's Runs.
- [Share a flow with a link](https://zerowidth.ai/docs/workbench/guides/share-via-url): Hand someone a URL and let them chat with your flow — no account, no setup on their end.
- [Put a flow in Slack](https://zerowidth.ai/docs/workbench/guides/share-via-slack): Connect a flow to a Slack channel so your team talks to it where they already work.
- [Put a flow in SMS](https://zerowidth.ai/docs/workbench/guides/put-a-flow-in-sms): Let anyone text your flow from any phone — no app, no account, no link to click.
- [Troubleshooting](https://zerowidth.ai/docs/workbench/troubleshooting): Diagnose flow runs — what the statuses mean, timeouts, test-pane-vs-API drift, and where to look.
- [SDK](https://zerowidth.ai/docs/workbench/sdk): Run a flow your team designed inside your own systems, instead of calling ours.
- [Workbench API reference](https://zerowidth.ai/docs/workbench/api): Endpoints for running Workbench flows programmatically.
- [POST /1.0/flows/:flowUuid/runs](https://zerowidth.ai/docs/workbench/api/run-flow): Execute a Workbench flow synchronously or as a stream.
- [GET /1.0/flows/public](https://zerowidth.ai/docs/workbench/api/list-public-flows): List public Workbench flows, newest activity first.
- [GET /1.0/flows/public/:flowUuid](https://zerowidth.ai/docs/workbench/api/get-public-flow): Read one public Workbench flow's metadata and full definition.
- [POST /1.0/flows/:flowUuid/fork](https://zerowidth.ai/docs/workbench/api/fork-flow): Fork a public Workbench flow into your workspace as a new, editable flow.
- [GET /1.0/tasks](https://zerowidth.ai/docs/workbench/api/list-tasks): List the tasks in your workspace, to find the one you want to run.
- [POST /1.0/tasks/:taskId/runs](https://zerowidth.ai/docs/workbench/api/start-task-run): Start a run of a task from your own code.
- [GET /1.0/tasks/:taskId/runs/:runId](https://zerowidth.ai/docs/workbench/api/get-task-run): Check in on a task run — where it got to, and what it's waiting for.
- [GET /1.0/knowledge-bases](https://zerowidth.ai/docs/workbench/api/list-knowledge-bases): List every knowledge base in the key's workspace.
- [POST /1.0/knowledge-bases](https://zerowidth.ai/docs/workbench/api/create-knowledge-base): Create an empty knowledge base.
- [GET /1.0/knowledge-bases/:id](https://zerowidth.ai/docs/workbench/api/get-knowledge-base): Read one knowledge base in full.
- [POST /1.0/knowledge-bases/:id/documents](https://zerowidth.ai/docs/workbench/api/add-knowledge-base-documents): Add documents to a knowledge base and start ingestion.
- [POST /1.0/knowledge-bases/:id/search](https://zerowidth.ai/docs/workbench/api/search-knowledge-base): Retrieve the most relevant chunks from a knowledge base.
- [GET /1.0/shims](https://zerowidth.ai/docs/workbench/api/list-shims): List every shim in the key's workspace.
- [GET /1.0/shims/:id](https://zerowidth.ai/docs/workbench/api/get-shim): Read one shim: its question and answers.
- [GET /1.0/shims/:id/weights](https://zerowidth.ai/docs/workbench/api/get-shim-weights): Fetch the published version's weights file.

## Compass

- [Compass](https://zerowidth.ai/docs/compass): Map how your business actually runs — and surface the changes worth trying.
- [Quickstart](https://zerowidth.ai/docs/compass/quickstart): Turn source material you already have into a mapped workflow with a change worth trying — in about ten minutes.
- [The map](https://zerowidth.ai/docs/compass/map): Your pages as a place you can look at, instead of a list you have to read.
- [Stakeholder interviews](https://zerowidth.ai/docs/compass/interviews): Pull the tacit knowledge out of the people who hold it — and wire it into the map.
- [Inbox](https://zerowidth.ai/docs/compass/inbox): Everything in Compass waiting on a human decision, in one queue.
- [Extract a map from a document](https://zerowidth.ai/docs/compass/guides/extract-a-map): Turn source material you already have into mapped pages, instead of building the graph by hand.
- [Run a stakeholder interview](https://zerowidth.ai/docs/compass/guides/run-an-interview): Pull the unwritten rules out of the person who holds them, and get them into the map.
- [Compass — programmatic access](https://zerowidth.ai/docs/compass/api): Read and build the Compass knowledge graph from an LLM client via the MCP server.

## Ledger

- [Ledger](https://zerowidth.ai/docs/ledger): The workspace decision memory — record what you did, what you expected, and whether you were right.
- [Quickstart](https://zerowidth.ai/docs/ledger/quickstart): Record a decision with its expectation, settle it against evidence, and watch the lessons start to compound.
- [Entries & beliefs](https://zerowidth.ai/docs/ledger/entries): The entry lifecycle in depth — kinds, expectations, evidence, settlement, and the rules that keep the record honest.
- [Plans and the calendar](https://zerowidth.ai/docs/ledger/calendar): Lay out the work that carries a decision out, see the quarter at a glance, and take it with you.
- [Metrics](https://zerowidth.ai/docs/ledger/metrics): The numbers the workspace watches — and the three ways readings get in.
- [Run a decision review](https://zerowidth.ai/docs/ledger/guides/decision-review): A recurring pass over open expectations — because pre-registration only pays off if settlement actually happens.
- [Ledger + Zero](https://zerowidth.ai/docs/ledger/zv1): Record decisions from conversation, recall them later, and let the assistant consult the record before acting.
- [SDK](https://zerowidth.ai/docs/ledger/sdk): Send Ledger readings from wherever your numbers already live, like a cron job, a webhook, a warehouse export or your own product.

## Napkin

- [Napkin](https://zerowidth.ai/docs/napkin): The suite's shared sketchbook — think with a pen, together, with an AI that can see the board.
- [Quickstart](https://zerowidth.ai/docs/napkin/quickstart): Sketch an idea, pull a teammate onto the board, and ask Zero what it thinks — in about ten minutes.
- [Drawing & the canvas](https://zerowidth.ai/docs/napkin/drawing): The tool rail, ink, shapes, text, stickies, images, selection, and PNG export.
- [Docs & sheets](https://zerowidth.ai/docs/napkin/docs-sheets): Napkin's two flat surfaces — a markdown document and a small spreadsheet — for the writing and the numbers that don't want a canvas.
- [Diagrams](https://zerowidth.ai/docs/napkin/diagrams): Flowcharts you build by dragging. Connect boxes, and the layout arranges itself.
- [Decks](https://zerowidth.ai/docs/napkin/decks): Slide presentations on the Napkin canvas — slides, layouts, text and lists, arranging shapes, links, present mode with a laser pointer, comments on a spot, and PDF/PowerPoint/PNG export.
- [Interfaces](https://zerowidth.ai/docs/napkin/interfaces): Small screens people can click — describing one, editing the files, giving it access to a flow or a shim, publishing a version, sharing a link, and going back through the history.
- [Brand kits](https://zerowidth.ai/docs/napkin/brand): Write your brand down once, with its voice, colours, typography, logo, imagery and the reasons behind each rule, and start every deck, doc and interface in Napkin from it.
- [Reference cards & connectors](https://zerowidth.ai/docs/napkin/references): Drop live flows, pages, datasets, and decisions onto a sketch — and wire them with connectors that follow.
- [Multiplayer & share links](https://zerowidth.ai/docs/napkin/sharing): Drawing together — workspace teammates natively, anyone else through a revocable link.
- [Zero can see your sketches](https://zerowidth.ai/docs/napkin/zv1): Ask the assistant about a diagram and it looks at the board — arrows, handwriting, stickies, and all.

## Prism

- [Prism](https://zerowidth.ai/docs/prism): Explore an idea, then find out what people really think — studies, interviews, and analysis in one place.
- [Run your first study](https://zerowidth.ai/docs/prism/quickstart): From a blank study to real answers you can read — in about five minutes.
- [Fields](https://zerowidth.ai/docs/prism/fields): Start from one idea and branch it along axes you name, until you can see the whole space.
- [Desk research](https://zerowidth.ai/docs/prism/research): What's already known about an idea, with the sources it came from.
- [Interviews](https://zerowidth.ai/docs/prism/interviews): Sit down with people one-on-one — guided by an AI interviewer — and let the conversation feed your study.
- [Reaching people](https://zerowidth.ai/docs/prism/reaching-people): Share links, your own audience, screeners, quotas, and weighting — get the right people and a sample you can defend.
- [Synthetic panels](https://zerowidth.ai/docs/prism/synthetic-panels): Try a study on people drawn from US population data before you send it to anyone, and read their answers as a separate, labeled result.
- [Reading the results](https://zerowidth.ai/docs/prism/results): Themes, segments, cuts, reports, tags — and turning a finding into something your team keeps.