Tableau + AI glossary: 30 terms, defined plainly

Tableau people and AI people use a lot of the same words to mean different things. "Agent" is a Tableau product and a general AI idea. "Token" is a login credential in Tableau and a unit of text in an LLM. "Extension" means one thing in Tableau and another in your code editor.

This glossary is the short version of each term as it applies to Tableau work, written so you can skim it before a meeting or link it to a colleague. Tableau terms are checked against Tableau's official docs and linked to them. Where I've written a deeper guide, the definition links there too. Product names and licensing are as of October 2026.

Agent

In AI, an agent is a model that can take actions in a loop: read something, decide what to do, call a tool, look at the result, and keep going until the job is done. A chatbot answers; an agent acts. Claude Code, GitHub Copilot's agent mode and Tableau Next's agents all fit this description. Don't confuse the general idea with Tableau Agent, which is a specific Tableau feature.

Claude

Claude is the family of large language models made by Anthropic, and the assistant you reach at claude.ai and in the Claude desktop and mobile apps. For Tableau work it's used for drafting calcs, writing admin scripts, reviewing workbook XML and building extensions. I compare it with the other options in Best AI tools for Tableau developers.

Claude Code

Claude Code is Anthropic's agentic coding tool. It reads your project, edits files, runs commands and connects to tools through MCP, and it runs in the terminal, IDEs, a desktop app and the browser. It's what I use to build Tableau extensions and admin automation. Start with AI for Tableau developers.

Connected app

A connected app is a Tableau feature that lets an external application authenticate to Tableau with signed JSON Web Tokens (JWTs) instead of a user's password. It's commonly used for embedding and for server-to-server automation. When you wire an AI tool to Tableau for many users, a connected app or OAuth is usually a better fit than a shared personal access token.

Context window

The context window is how much text a model can consider at once: your prompt, any files you attach, the conversation so far and the model's reply. It's measured in tokens. It matters for Tableau because a large .twb file is a lot of XML, and if it doesn't fit, the model can only review part of it. See Review a Tableau workbook with AI.

Dashboard extension

A dashboard extension is a web app that runs inside a Tableau dashboard as its own object. It can read data from the dashboard's worksheets, apply filters and parameters, and show its own interface, such as a search box or a KPI card. It's described by a .trex manifest and built with the Extensions API. Walkthrough: Build a Tableau extension with Claude.

Einstein Trust Layer

The Einstein Trust Layer is Salesforce's set of security and governance controls that sits between Salesforce products and the large language models behind them. Tableau's AI features, including Tableau Agent and Tableau Pulse insight summaries, are built on it. If your security team asks how Tableau's built-in AI handles your data, this is the term to look up.

Embedding API

The Tableau Embedding API (v3) is a JavaScript library for putting Tableau views inside your own web pages, using web components such as <tableau-viz>. You can set display options in HTML and control the embedded viz with JavaScript, for example to apply filters. It's for embedding Tableau somewhere else, which is the opposite direction from the Extensions API.

Extensions API

The Tableau Extensions API is the JavaScript library that lets web code run inside Tableau and talk to it. It covers both dashboard extensions and viz extensions: reading summary or underlying data, listening for filter and selection changes, saving settings and opening configuration dialogs. AI assistants write this code well once they know its rules, which is what a skill is for.

Hallucination

A hallucination is when a model states something false with full confidence. In Tableau work the common ones are invented field names, functions that don't exist in Tableau's calc language, and REST API endpoints or parameters that were never real. The fix is to give the model ground truth (your field list, the actual error, the official docs) and to test everything it writes before you trust it.

LLM (large language model)

An LLM is a model trained on large amounts of text to predict and generate language. Claude, the GPT models behind ChatGPT, and the models behind Tableau's AI features are all LLMs. On their own they only know their training data; tools, MCP and RAG are how they get access to your current Tableau content.

LOD expression

A level of detail (LOD) expression is a Tableau calculation that computes an aggregate at a granularity you specify, using FIXED, INCLUDE or EXCLUDE, independent of the view's level of detail. LODs are one of the places AI drafting helps most, because the syntax is short but the logic is easy to get wrong. Tell the model the grain of your data and what one row means, then check the result against a known number. More in Tableau calculated fields with AI.

MCP (Model Context Protocol)

MCP is an open standard for connecting AI applications to outside systems: data sources, tools and workflows. Anthropic introduced it, and it's now supported by Claude, ChatGPT, VS Code and many other clients. For Tableau, it's the standard way to let an AI assistant query your sites without copy and paste. See Tableau MCP server with Claude.

MCP server

An MCP server is the program on the other end of the protocol. It exposes a set of tools (and sometimes resources and prompts) that an AI client can call. Tableau MCP is one; the TableauOps MCP connector is another, which lets your AI publish and host viz extensions. A server can run locally on your machine or remotely over HTTPS.

Metadata API

The Tableau Metadata API is a GraphQL API that catalogs content on Tableau Cloud and Tableau Server: workbooks, data sources, flows, fields and the lineage between them. It answers questions like "which workbooks use this published data source" in a single query. It's a great source of ground truth to hand an AI for documentation and impact analysis.

Personal access token

A personal access token (PAT) is a long-lived credential for signing in to the Tableau REST API without a username and password. Per Tableau's docs, a PAT expires if it goes unused for 15 consecutive days, and site admins set its maximum lifetime. Signing in again with the same PAT ends the previous session, so give each script or AI tool its own token.

Prompt

A prompt is the input you give a model: your question, instructions and any context. Most bad AI output for Tableau comes from prompts that leave out the context (field names, data grain, Tableau version, the exact error). I keep a set that works in Prompts for Tableau viz extensions.

RAG (retrieval-augmented generation)

RAG is a pattern where a system first retrieves relevant documents or data, then hands them to the model along with your question so the answer is grounded in that material. It's how an assistant answers from your own docs or catalog instead of its training data. Tableau MCP's metadata tools and the Metadata API are common retrieval sources for Tableau.

REST API

The Tableau REST API is the HTTP interface for managing Tableau Cloud and Tableau Server: users, groups, projects, permissions, publishing, extract refreshes and more. Most admin automation goes through it, often via TableauServerClient. AI is good at writing these scripts and better at fixing them when it can run them. See Tableau REST API automation with Claude.

Skill (Claude)

A skill is a folder with a SKILL.md file of instructions, plus any reference files, that Claude loads when it's relevant to the task. It's how you give the model your rules once instead of re-explaining them every session. Skills follow the open Agent Skills standard. Mine for Tableau extensions is free here, and the deeper guide is Claude Code skills for Tableau.

Tableau Agent

Tableau Agent is Tableau's built-in conversational AI assistant. It can build visualizations from natural language, suggest questions, and create, explain and edit calculated fields. On Tableau Cloud it requires Tableau+ with AI in Tableau turned on; it's also available on Tableau Server 2025.3 and later, and in Tableau Desktop 2025.1 and later when signed in to a supported site.

Tableau MCP

Tableau MCP is Tableau's official, open-source MCP server. It gives AI clients tools such as list-datasources, get-datasource-metadata, query-datasource, list-workbooks and get-view-image, working against Tableau Cloud or Tableau Server with the user's permissions. Tableau also documents a hosted version for Tableau Cloud customers. Setup: Tableau MCP server with Claude.

Tableau Next

Tableau Next is Salesforce's agentic analytics platform, built on the Salesforce platform and integrated with Agentforce. It's a separate product from classic Tableau Cloud, Server and Desktop, aimed at AI agents working alongside people across data prep, semantics and analysis. It's sold with the Tableau+ SKU.

Tableau Pulse

Tableau Pulse is a metrics and monitoring layer for Tableau Cloud. You define a metric, people follow it, and Pulse automatically detects trends, drivers and outliers and writes AI-generated summaries of what changed. It's Tableau Cloud only; base features are broadly available and premium capabilities need Tableau+.

Tableau+

Tableau+ is the premium license edition of Tableau Cloud, sold per user by role (Creator, Explorer, Viewer). It's the edition that unlocks Tableau's AI features on Cloud, including Tableau Agent and premium Pulse capabilities. If someone says "Tableau AI isn't on for us," missing Tableau+ is the first thing to check.

TableauServerClient (TSC)

TableauServerClient is Tableau's official Python library for the REST API. It wraps sign-in, pagination and common objects so scripts are shorter and easier to read. Because it's so widely used, AI models usually know it well, which makes it a good default language for AI-written admin scripts.

Token

In AI, a token is a chunk of text (often part of a word) that models read and write; context windows and API usage are measured in tokens. In Tableau, "token" usually means an authentication credential, such as a personal access token or a REST API session token. When an AI tool asks for "your Tableau token," it means the credential, so treat it like a password.

.trex manifest

A .trex file is the XML manifest that registers an extension with Tableau. It holds the extension's ID and version, the HTTPS URL where it's hosted, the permissions it needs, author details and the minimum API version; for viz extensions it also declares the encodings that appear on the Marks card. Tableau documents both the dashboard manifest and the viz manifest. A missing required element is the most common reason a freshly AI-built extension won't load.

Viz extension

A viz extension adds a new mark type to the Marks card on a Tableau worksheet, so you can build a chart Tableau doesn't have natively (a Sankey, a ridgeline, a chord diagram) by dropping fields onto encodings you declare. Viz extensions arrived in Tableau 2024.2 and use the same Extensions API. You can see working ones in the TableauOps extensions gallery and build your own in Studio.

VizQL Data Service

VizQL Data Service (VDS) is a Tableau API for querying published data sources over HTTP without building a viz. You describe the query as JSON and get data or metadata back. It's available on Tableau Cloud and on Tableau Server 2025.1 and later, and it's what Tableau MCP's query-datasource tool uses to answer questions from governed data.

Keep going

The Tableau + AI hub groups every guide by what you're trying to do. If you're choosing tools, start with Best AI tools for Tableau developers. If you want to see AI applied to Tableau operations, look at Scout.

Eric Summers, 2x Tableau Ambassador, Tableau Technology Partner, and founder of TableauOps