Open Pixel Agents
Forked from the original pixel-agents project, this version is maintained as an evolving toolkit that supports both direct AI models and OpenCode sessions.
Open Pixel Agents is a VS Code extension that turns your AI Agents and OpenCode sessions into animated pixel characters inside a virtual office. Each session appears as a character that can walk, sit at a desk, show status bubbles, and reflect current activity in real time.
Current Features
- Multi-Provider AI Support — the extension allows you to select from multiple AI providers directly inside the modal (OpenAI, Claude, Qwen, OpenClaw, and more).
- OpenCode Runtime — you can also launch and visualize OpenCode sessions through the local OpenCode server API and SSE event stream, mapping them seamlessly to your VS Code terminals.
- One session, one character — each primary session or task becomes a pixel agent in the office.
- Sub-agent visualization — Agent tool use, child sessions, or subtasks appear as temporary sub-agent characters.
- Live status feedback — characters react to working, reading, waiting, permission requests, retries, and completion.
- Workspace session persistence — reopening the project restores your AI Agents mapping to the UI layout.
- Seat assignment and office simulation — agents can be selected, reassigned to seats, and return to their desks automatically.
- Layout editor — edit floors, walls, furniture placement, and office arrangement directly in the webview.
- Import/export layouts — save and share office layouts as JSON.
- Optional completion sound — play a chime when work completes.
- Mouse camera controls — left-drag to pan, middle click to re-center, wheel to zoom.
Requirements
- VS Code 1.50.0 or later
- For the local "VS Code Terminal" provider, the OpenCode CLI is required to be installed globally and available on your
PATH.
- For other AI providers (OpenAI, Claude, Qwen, OpenClaw), an appropriate API key and/or local model server is required.
Getting Started
Run from source
git clone https://github.com/inkbottle/open-pixel-agents.git
cd open-pixel-agents
npm install
cd webview-ui && npm install && cd ..
npm run build
Then press F5 in VS Code to launch the Extension Development Host.
Basic usage
- Open the Open Pixel Agents panel
- Click + Agent to create a new Agent.
- Select an AI Provider (such as OpenAI, Claude, Qwen, or OpenClaw) and provide your model and key, OR select "VS Code Terminal" to launch an OpenCode session.
- Work in the attached terminal (for OpenCode) or interact directly while the office reflects session and subtask activity in real time.
- Click an agent to focus/select it.
- Click a seat to reassign the selected agent.
- Click the close button on an agent to remove it.
How It Works
When you create an agent via an AI API provider, the extension connects to the configured model's stream, spawning characters for tool calls or processing steps.
When you create a "VS Code Terminal" agent, the extension:
- Ensures a window-local OpenCode server is available (port auto-resolved).
- Creates a new OpenCode session over HTTP.
- Opens a VS Code terminal and runs:
opencode attach http://127.0.0.1:<resolved-port> --session <session-id> --dir "<workspace>"
- Subscribes to OpenCode's
/global/event SSE stream.
Layout Editor
The built-in editor supports:
- floor painting
- wall painting
- erase/select/pick/place tools
- furniture move / rotate / delete
- undo / redo
- layout import / export
- expandable grid editing up to 64×64
Office Assets
The office tileset used by the optional asset pipeline is Office Interior Tileset (16x16) by Donarg.
This tileset is not included in the repository. To use the full furniture catalog locally, purchase the tileset and run:
npm run import-tileset
The extension still works without the tileset using the bundled default assets and layout.
Development
npm install
cd webview-ui && npm install && cd ..
npm run build
Package the extension with:
npx @vscode/vsce package
Authors & Maintainers
Buy Me a Coffee
If this extension has been helpful to you, a coffee would be greatly appreciated
License
This project is licensed under the MIT License.
