[English] | 简体中文

CharDesk

A shared visual medium for people and agents.

CharDesk gives people and LLM-powered agents a common artifact: visual Unicode text. Its infinite Canvas lets people draw diagrams and interfaces while agents inspect the source, revise it, and verify the result. Cell UI applies the same medium to interactive interfaces.

Open Canvas · Explore Cell UI

Documentation · Product home · CharGraph

Start with your agent

Requires Node.js 20 or later.

npx skills add https://github.com/sayhi-bzb/chardesk --skill chardesk

Register the MCP server with your agent:

# Codex
codex mcp add chardesk -- npx -y @chardesk/mcp

# Claude Code
claude mcp add chardesk -- npx -y @chardesk/mcp

# Pi
pi mcp add chardesk -- npx -y @chardesk/mcp

After that, no manual server or pair command is needed. The first Canvas tool call opens Canvas when necessary, and the page connects to the local bridge automatically.

Then tell your agent what you want to see:

$chardesk Explain how a GPU works as a visual Canvas.

The agent creates the source, checks it, and opens the canvas. You do not need to learn a file format or configure an AI provider first.

What you can make

• Explain a concept as a spatial story instead of a wall of text.
• Present an idea as character-based slides.
• Build scientific figures with plots, formulas, data, and annotations.
• Sketch interfaces, terminals, dashboards, and product states.
• Map architectures, workflows, timelines, and relationships.
• Keep a Canvas that people and agents can inspect and revise together.

Shared medium · Source

Visual explanation · Source

Scientific figure · Source

Character slides · Source

Interface design · Source

Agent Canvas · Source

ANSI interface studies

Context Control Room · Source

Idea Signal Player · Source

Pocket Canvas · Source

Why text can be visual

People scan a two-dimensional surface. Language models generate and edit token sequences. Screenshots preserve layout, but add pixel noise and are awkward to revise precisely across turns; plain text is easy to edit, but normally gives up space and style.

CharDesk keeps both. A fixed Unicode grid carries position, box drawing carries structure, and ANSI carries emphasis. The result remains selectable, searchable, diffable, and directly editable by an agent.

   Person reads and uses
           ⇅
   Shared Unicode scene
           ⇅
   Agent reads and edits

The medium

Visual text

Unicode, box drawing, CJK, technical symbols, monochrome emoji, and Nerd Font glyphs share one grid. ESC-less ANSI adds foreground and background colors, bold, italic, underline, strike, and inverse styles without turning the work into an image.

Structured expression