Skip to main content
DocumentationLibraryReference
On this page

Getting Started

What is Kater?

Kater does for analytics what AI coding tools did for software engineering: AI does the work, and every change gets checked before anyone relies on it.

Business users build their own apps and get answers in chat, Excel, SQL or any tool they already use, all on metrics that data teams govern the way engineers govern code.

For business teams

  • Build in your agent, share through Kater. A dashboard you make in Claude or Codex becomes a Kater app. Each figure cites its cell in the app’s workspace, where you see how the number is made.
  • One shared space. Share the app from Kater, not as a copy. Business teams and the data team work on the same app and workspace.
  • Define what you need. Write your own measure, and propose it to the data team when it proves useful.

For data teams

  • Your guardrails, under every app. People already build dashboards with agents. Kater makes the SQL and sources under them match your definitions.
  • Semantic Modeling as code. Kater SQL is your warehouse’s own SQL with a few additions. Even a measure with CTEs is reusable anywhere.
  • Verify and promote. Tests keep verified numbers correct. Kater finds the same formula in several workspaces, and you approve it for the shared layer.
  • Everything in git. Models, queries, dashboards and apps are files. Work on a local copy, and write the shared layer back to your own GitHub repository.

Start here

Each person starts with one prompt or one link.

Set up Kater For the admin. Start your free trial, then give one prompt to your agent, which sets up Kater with you. Set up Kater
Start as a Developer For a Developer whom the admin invites. Your agent sets up your computer from the same prompt. Start as a Developer
Start with Kater For everyone else. Connect Claude or ChatGPT to Kater, and ask your first question. Start with Kater

Find the right starting point

Start with what you want to do. Each path points you to the guide that best matches your goal.

Ask your data a question Connect Claude, ChatGPT, Claude Code, Codex or Cursor to your organisation’s Kater MCP, then ask in plain English. Connect your agent
Build an app from a prompt Describe a report, a slide deck or an interactive page. Your agent builds it on governed queries, and you can export it. Build an app
Start from the Library Copy a ready-made app, starter or scheduled workflow, such as a monthly retention scorecard or a Monday deck for the CMO. Browse the Library
Turn a question into a reusable query Save an analysis as a query and a chart in a workspace, so your team can run it again and build on it. Explore workspaces
Connect your data Connect Snowflake and sync your schemas, through your agent, the web app or the CLI. Connect your warehouse
Define your metrics Model datasets and semantic models in Kater SQL, so every answer, app and dashboard uses one definition. Learn how modeling works
Check a number Ask your agent to trace any figure back to its source in the warehouse, step by step. Use the starter

Explore products

Business teams

  • Your AI data assistant. Connect your own agent to your organisation’s Kater MCP and ask in plain English. Every figure comes from a query that runs on your warehouse, through definitions your data team governs.
  • Starters. Ready-made prompts such as Brief me, Compare two periods and Explain a metric, in the Ask panel and in your agent.
  • Scheduled workflows. Workflows that your own agent runs on a schedule, such as a brief every Monday at 08:00, in Claude Cowork, Codex or Cursor.
  • Apps. Reports, slide decks, chat threads and interactive pages that your agent builds on live queries. Export any page to PowerPoint, PDF, Word, Markdown or Excel.
  • Dashboards and Explore. A dashboard for every workspace, and a board of ready-made cards for every shared model.

When you ask Why did it change?, your assistant runs a complete analysis workflow:

  1. It finds the workspace that answers the question.
  2. It writes a plan.
  3. It builds the queries the plan needs.
  4. It profiles the data and calibrates any thresholds.
  5. It runs the queries and finds the drivers.
  6. A second agent reviews the work in a fresh context.
  7. It reports the answer in a note in your brainstorm folder.

Data teams

Platform admins

  • Connections. Connect Snowflake and sync its schemas, through your agent, the admin console or the CLI.
  • Organisation and people. Sign-in, roles and groups through PropelAuth, with SSO and SCIM for your IT team.
  • Access. Drive-style sharing on folders and files, with Viewer, Editor and Manager levels.
  • Content store. Kater keeps your content. You can write the shared layer back to your own GitHub repository.
  • Admin console. Who uses Kater and what needs attention, every object, promotion requests, users and permissions.

Understand how Kater works

Before setting up, these pages help you build a mental model of Kater:

  • Database datasets: one generated file for each warehouse table, written by schema sync. They are the source layer.
  • Kater datasets: the governed view of one table. Each column becomes a named dimension, and measures are defined once.
  • Semantic models: Kater datasets joined into the model people ask questions of. With Kater datasets, they make up the shared layer.
  • Workspaces: the boundary for a project or a team. A workspace reads one shared model and holds queries, charts, a dashboard, tests and apps. It is personal, or it belongs to the connection.
  • Queries: reusable Kater SQL saved in a workspace, which compiles to governed warehouse SQL. Charts, dashboards, apps and agents all read queries.
  • Apps: HTML pages, reports and decks that draw live query results through Kater’s tags.
  • The Kater MCP: the MCP server Kater hosts for your organisation. Every agent connects to it and acts as the person who signed in, with that person’s access.
  • Starters, skills and workflows: Markdown files that teach your assistant how to brief, analyse, build and check. Your data team edits the shared ones, and anyone can add their own. A workflow with a schedule runs on its own in your agent.
  • Promotion requests: how a definition that proved useful in one workspace moves into the shared layer for everyone.

Our philosophy

Kater is built by data engineers and analysts with a decade of working with business teams. These are core principles which, through trial and tribulation, they’ve distilled as non-negotiables for teams that want to take meaningful action with their data.

  1. No vendor lock-in.

    At the end of the day, you should own everything. Your definitions are plain SQL and Markdown files. Clone them with kater git clone, read them, and change them with the tools you already use. Write the shared layer back to your own GitHub repository whenever you want. No black boxes that abstract away your business metrics.

  2. A metric should mean the same thing in every meeting.

    ROAS in the QBR deck, ROAS on the dashboard and ROAS the CFO quotes should be one definition, not three. Define it once in the shared layer and reuse it everywhere. When it changes in one place, every query, app and agent answer that reads it changes with it.

  3. Answers should move at the speed of the business.

    A question asked on Monday morning should not wait in a ticket queue. Anyone can ask their own agent and get an answer from governed definitions while the question still matters. Starters do the routine work: a brief of what moved, a comparison of two periods, a trace of why a number changed.

  4. Every action you take should inform the next decision.

    Today, the loop is broken: data surfaces an insight, someone acts on it, and the story ends. In Kater, work leaves a trail you can build on. Plans, checks and findings stay as notes in your brainstorm folder. Every workspace and app keeps its History, and you can undo any change. Promotion turns a definition one team proved into the one everyone uses.

The verification loop

As people work with their assistants, Kater gives data teams a feedback loop to watch and verify what the agents use, so answers stay accurate as usage grows.

  1. The agent answers from governed fields.

    It runs queries in a workspace over the shared layer. Figures come only from those runs, never from the agent’s own arithmetic.

  2. The server decides what is verified.

    A field is verified only when a shared model under modeling/ defines it. A field that a workspace or a query computes is not. The field list marks which is which.

  3. The data team sees what repeats.

    Promotion recommendations find the same formula in several workspaces. People also send promotion requests from a workspace’s Promote sidebar, with their reason.

  4. The data team approves.

    Each draft gets a formula check and a check of its results on the warehouse before the data team approves it in the admin console.

  5. The field moves into the shared layer.

    The requester’s definition is replaced by a reference to the shared field.

  6. Everyone’s answers use it.

    Every workspace, app and agent that reads the model now reads one verified definition.

Read more in Promotion and verification and Promotion requests.