DocsBack to homepage

Start Here

  • Getting Started
  • Key Concepts

Design Tokens

  • Token Types
  • Token Modes
  • Token Enforcement
  • Deprecated Tokens
  • Quality and Accessibility

Components

  • Component Builder
  • Composition Rules

Publishing

  • Publishing
  • Docs Mode
  • Changelog Notifications
  • Notifications and Alerts

Integrations

CLI & Data

  • CLI Reference
  • CLI Configuration
  • Import Formats
  • Importing Tokens
  • Export Formats

Tooling

  • Studio AI Assistant
  • Figma Plugin
  • API Reference
  • Webhooks

Account & Billing

  • Audit Log
  • Security and Access
  • Account Security
  • Pricing and Payments

Documentation

Studio AI Assistant

The Studio AI assistant is a conversational panel built into the Studio. Ask it questions about your design system, propose token changes, manage branches, or rename settings. Everything happens through plain language, and nothing changes until you approve it. Available on Pro plans.

Opening the assistant

Press Command-/ on Mac or Ctrl+/ on Windows from anywhere in the Studio. You can also open it from the TopBar.

When the conversation is empty, the panel shows four suggested prompts to help you get started.

Plan requirement

The AI assistant is a Pro plan feature. Free plan members see an upgrade prompt in place of a response. Upgrade to Pro

Rate limits

Pro workspaces can send up to 50 messages per day. The counter resets at midnight UTC.

When the limit is reached, the assistant shows a message inside the panel. Wait until the counter resets before sending more requests.

What the assistant can do

The assistant handles five types of requests. Each one is described below with what triggers it and what you see in the panel.

Answer questions

Ask about concepts like token aliases, modes, branches, or component slots and the assistant gives a direct explanation. It uses context from your current workspace and project settings, so answers reflect your actual setup.

Navigate the docs

When you ask a how-to question, the assistant may open the relevant docs section directly inside the panel. Supported sections: Getting Started, Tokens, Aliases, Modes, Components, and Branches.

Propose a token or component change

When you ask the assistant to suggest or apply a change (for example, "update my primary color to match my button background"), it renders a proposal card in the chat showing the current value, the proposed value, and the reason for the change.

Nothing updates until you accept the proposal. Rejected proposals stay in the conversation history so you can refer back to them.

Manage branches

When you ask to create a branch, merge the current branch to main, or delete a branch, the assistant renders a branch proposal card with the action and target.

Nothing executes until you approve. This makes the assistant useful as a confirmation step before merging to your main design system.

Rename project settings

When you ask to rename the workspace, a dimension, or a mode, the assistant proposes the rename in a card before applying it.

The assistant cannot change billing settings, org permissions, workspace slugs, or token prefixes.

What the assistant cannot do

  • •It does not edit tokens or branches without showing a proposal card first. Every change requires your explicit approval.
  • •It cannot access data outside your current workspace. There is no cross-org or cross-workspace context.
  • •It cannot change billing settings, org permissions, workspace slugs, or prefix values.
  • •It does not retain conversation history between sessions. Each time you open the panel, the conversation starts fresh.

Tips for effective use

  • •Be specific about what you want changed. Include the token name and the target value when asking for a change.
  • •Ask for an explanation before a proposal. "Why is my primary button color overriding the semantic alias?" gives you useful context first.
  • •Use it for branch management when you want a built-in confirmation step before merging.
  • •If you are close to the daily limit, use it for complex questions rather than quick lookups.