> ## Documentation Index
> Fetch the complete documentation index at: https://docs.myrmagent.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Canvas Workspace

> tldraw-powered infinite canvas for visual brainstorming, Agent-driven diagram building, and real-time collaboration between you and your AI Agent.

# Canvas Workspace

Myrm includes a built-in **infinite canvas** powered by [tldraw](https://tldraw.dev), giving you a freeform visual workspace for brainstorming, diagramming, and visual prompting — all without leaving the platform.

## How It Works

<Steps>
  <Step title="Create a Canvas">
    Navigate to **Canvas** from the sidebar menu. Click **New Canvas** to create a fresh workspace. You can manage multiple canvases — rename, switch between, or delete them at any time.
  </Step>

  <Step title="Draw and Annotate">
    The full tldraw toolkit is available: freehand drawing, shapes, text, arrows, sticky notes, and more. Everything auto-saves with debounced persistence — no manual saving needed.
  </Step>

  <Step title="Agent Interaction">
    Your AI Agent can read the canvas state, see what you've selected, and insert new elements. This enables **visual prompting** — draw a rough sketch, select it, and ask the Agent to refine or implement it.
  </Step>

  <Step title="Real-Time Sync">
    When the Agent modifies the canvas (via MCP tools), changes appear instantly through SSE (Server-Sent Events). Multiple browser tabs stay in sync.
  </Step>
</Steps>

## Enabling Canvas Tools

To let the Agent interact with your canvas:

1. Open the **chat config panel** (click the expand button next to the send button)
2. Click **Built-in Tools**
3. Toggle **Infinite Canvas** (无限画布) on
4. Navigate to or open a canvas — the Agent will automatically bind to the active canvas

Once enabled, the Agent gains three canvas operations:

| Tool                        | What It Does                                                              |
| --------------------------- | ------------------------------------------------------------------------- |
| **canvas\_get\_state**      | Reads the full tldraw snapshot — the Agent sees everything on your canvas |
| **canvas\_get\_selection**  | Reads which shapes you've currently selected — the Agent knows your focus |
| **canvas\_insert\_element** | Adds new shapes (text, notes, rectangles, arrows, etc.) to the canvas     |

This enables workflows like:

* Draw a wireframe → ask the Agent to generate code from it
* Select a diagram → ask the Agent to explain or improve it
* Ask the Agent to visualize a concept → it draws directly on your canvas
* Paste meeting notes → ask the Agent to create a mind map on your canvas

## Multi-Canvas Management

Unlike single-canvas tools, Myrm supports **multiple independent canvases**:

* Create canvases for different projects or brainstorming sessions
* Rename canvases for easy identification
* Delete canvases you no longer need
* Canvas metadata is stored in SQLite; snapshots are persisted as JSON files

## Security

* **Path traversal protection**: Canvas IDs are validated as UUIDs before any filesystem operation
* **Non-blocking I/O**: All file reads/writes use `asyncio.to_thread` to avoid blocking the event loop
* **SSE reconnection**: Automatic 3-second backoff reconnection if the event stream disconnects

## Internationalization

The canvas UI is fully localized in **English** and **Chinese**, including all labels, buttons, empty states, and confirmation dialogs.