Skip to main content

Using AI

AI coding tools like Claude Code, GitHub Copilot and Codex can write UDFs, wire widgets and push canvases for you. Fused publishes its skills plugin so the tool already knows the canvas.toml format, how to structure a UDF, which props each widget takes and which fused CLI commands to run. You don't have to explain any of it in your prompts.

To build UDFs that agents call as tools, see Building for Agents.

Use Fused Render​

Fused Render is a desktop app that lists every canvas in your Workbench account. Sign in, open a canvas, and edit it directly or ask a coding agent on your machine to change it: Claude Code, GitHub Copilot, Codex or Gemini, signed in with your own subscription. Every save is published back to Workbench, so there is nothing to pull or push. See Fused Render to install it and turn on Canvases.

To use a coding tool in your own terminal instead, install the Fused skills below.

Install the Fused skills​

Install the workbench plugin in your coding tool. The skills use the fused CLI to push and run canvases, and when it's missing they ask your tool to install it with uv, so you don't need it first. To use the fused workbench ... plugin add shortcut below, install it yourself:

uv tool install fused

To install with pip instead, see Install.

Install Claude Code, then:

fused workbench claude plugin add

Or from Claude Code directly:

claude plugin marketplace add fusedio/skills
claude plugin install workbench@fused-marketplace

Update with fused workbench claude plugin update and remove with fused workbench claude plugin remove.

The first time the tool runs a Fused command, your browser opens so you can log in to Fused. Finish logging in within 30 seconds, or run fused workbench login yourself.

Install the skills without the plugin

If your organization turns off plugins, install the skills on their own. They're plain Agent Skills folders. Clone them once:

git clone https://github.com/fusedio/skills ~/.fused-plugins

Then link them where your tool reads personal skills. The Copilot CLI and VS Code read ~/.agents/skills, and Claude Code reads ~/.claude/skills:

mkdir -p ~/.agents/skills ~/.claude/skills
ln -s ~/.fused-plugins/workbench/skills/* ~/.agents/skills/
ln -s ~/.fused-plugins/workbench/skills/* ~/.claude/skills/

To update them, run git -C ~/.fused-plugins pull.

Try it out​

Ask your coding tool to build a canvas from scratch:

Make a Hello World canvas with some dummy data and a chart showing the number of rows in the data

It writes the UDF, wires up a chart widget and pushes the canvas to your account with the fused CLI:

Hello World canvas with dummy data and a bar chart of row counts by category

Edit an existing canvas​

Pull the canvas to a local folder. This downloads the UDF .py files, canvas.toml and widget JSON into a folder named after the canvas:

fused workbench canvas pull <canvas_name>

Ask your coding tool to change the files, then push that folder back:

fused workbench canvas push <canvas_folder>

The push replaces the remote canvas with your local folder, so any UDF missing locally is removed. Workbench then shows that the server has a newer version:

Workbench showing newer version on server

Click Compare to review the diff, then pull all UDFs at once or pick individual ones:

Pull server changes panel with per-UDF controls

See Versioning for checkpoints, diffs and restoring earlier states.

AI Chat in Workbench​

Workbench also has a built-in AI Chat for quick questions and small edits in the browser.