Sayhi-bzb/CharDesk

What it solves

CharDesk provides a shared visual medium that allows humans and LLM agents to collaborate on visual content using Unicode text. It solves the problem of the gap between how humans perceive visual layouts (two-dimensional surfaces) and how LLMs generate and edit token sequences (linear text), allowing for content that is selectable, searchable, and directly editable by agents while remaining visually structured.

How it works

CharDesk uses a fixed Unicode grid combined with box drawing characters, CJK symbols, and ANSI styles for color and emphasis. It converts structured expressions—such as Markdown, Mermaid, math, and JSON—into portable character graphics via CharGraph. This content is delivered through an infinite canvas or interactive "Cell UI" interfaces. Agents interact with this medium through a Model Context Protocol (MCP) server, enabling them to read, search, and write to the active visual surface.

Who it’s for

• AI Agents: LLMs that need to create and maintain visual explanations, diagrams, and interfaces without generating images or complex file formats.
• Builders: Developers who want to integrate visual text rendering using the @chardesk/protocol, @chardesk/viewer, and @chardesk/fonts packages.
• Users: People who want to collaborate with agents to map architectures, sketch interfaces, and create scientific figures using a character-based medium.

Highlights

• Shared Visual Medium: Uses Unicode and ANSI to create a visual language that both humans and agents can understand and and edit.
• MCP Integration: Native support for agents like Claude Code, Codex, and Pi via a local MCP server.
• Agent-Editable: Content remains as text, making it easy for agents to diff and revise across turns.
• Versatile Output: Supports the creation of spatial stories, character-based slides, scientific figures, and interface sketches. -n- Structured Expression: Compiles Markdown, Mermaid, and Vega-Lite into character graphics.