Cursor 3 Interactive Canvases: Live Dashboards Inside Your Editor

Cursor 3 shipped a feature on April 15 that changes what an AI coding session can output. Instead of just writing code or chat replies, the agent can now build an interactive canvas — a live dashboard, diagram, or visualization — that sits next to your terminal in the Agents Window. Canvases are durable artifacts. They stick around between turns, you can keep editing them, and they share state with the code the agent is working on.

Most developers we have talked to are still treating canvases like a fancier chat output. That misses the point. A canvas is a tool the agent can build for you on the fly when text is the wrong format for the answer.

What a Canvas Actually Is

A canvas is a panel that lives alongside your code editor in the Agents Window. It can contain tables, charts, diagrams, to-do lists, code diffs, and custom interface boxes. The agent generates the canvas using component primitives Cursor provides, so the output is interactive — clickable, sortable, drillable — not a static screenshot.

The point is that some answers are not text. If you ask the agent to map out the dependencies between 12 modules in your codebase, a list is hard to read. A force-directed graph shows the structure in two seconds. The canvas system gives the agent a way to pick the right format for the question.

Three Workflows That Make Canvases Worth It

After two weeks of trying canvases on real work, three patterns stood out. Each one solves a problem that text-based output handles poorly.

Workflow 1: Codebase Maps for Onboarding

Drop into a new repo and ask the agent to build a canvas mapping the main modules and how they connect. You get a clickable diagram with module names as nodes and import relationships as edges. Click any node and the agent expands it to show the public API of that module. This is faster than reading a README and more accurate than the README usually is.

Build a canvas that shows the main modules in src/ and how they import each other. Make each module clickable so I can see its public exports. Skip test files and node_modules. Use a force-directed layout.

Workflow 2: Test Failure Dashboards

When a test run fails with 30 failures, you have to hunt through terminal output. Ask the agent to build a canvas dashboard from the failures: a sortable table grouped by file, with the assertion message, expected vs actual values, and a button to jump to the failing line. The agent updates this canvas after each test re-run, so you can watch failures drop as you fix them.

Run the test suite. Build a canvas with a table of all failures. Columns: file, test name, assertion message, expected, actual. Group rows by file. Add a button on each row that opens the failing line in the editor.

Workflow 3: Migration Progress Trackers

Doing a big migration — switching ORMs, swapping a logging library, moving from REST to GraphQL? Ask the agent to build a canvas that tracks every file touched by the migration, its current state (untouched, in progress, migrated, verified), and the percent complete. Pin it to the side panel and the agent updates it as work proceeds. You always know where you are without scrolling chat history.

How Canvases Compare to Notebooks

Jupyter notebooks have always offered the same idea: code and visual output side by side. The difference is that a canvas is built by the agent on demand, lives in your existing editor, and updates as the agent does work. You do not need to switch contexts to a separate notebook environment, and you do not need to write the visualization code yourself.

If you already work in notebooks, canvases will not replace them. They are better for ephemeral, in-the-moment views — the kind of dashboard you would never bother to write but would benefit from having for the next ten minutes.

Setup and Limitations

Canvases require Cursor 3 or later. Open the Agents Window (cmd+shift+A on Mac), start an agent task, and ask for a canvas in your prompt. The canvas appears in the side panel. You can pin it, close it, or have multiple canvases open at once.

Canvases do not export cleanly to other formats yet. If you want to keep one as documentation, take a screenshot or ask the agent to convert it to a markdown table. Cursor has hinted at export-to-Notion and export-to-PDF for a future release, but as of April 2026, neither exists.