Coding Agent

Formerly known as Baize (白泽).

First time here? Start with the Quick Start (download the code → install dependencies → run the desktop app → pair a browser).

A locally runnable AI coding assistant for chat, reading and editing code, tool use, and browser automation. Supports the desktop app, Web UI, and VS Code / Cursor / IntelliJ integrations.

How do you want to use it?

Core capabilities

• Coding assistance — Read and edit files, run shell commands, search the codebase, and explore in parallel with subagents
• Browser automation — Open web pages, click controls, fill forms, take screenshots, and read page content; can control your own Chrome profile for sites that require authentication
• Extensibility — Skills, Commands, Plugins, MCP, and custom Subagents
• Multiple interfaces — Electron desktop app, Web UI, ACP (IDE sidebar), and HTTP API / SDK
• Memory and compaction — Session memory, Auto memory, and automatic context compaction

Five-minute desktop quick start

Prerequisites: Node.js 20+ and API configuration supplied by your team (see .ai-agent/settings.json)

git clone <your-repository-url>
cd coding-agent
npm install   # root + workspaces (web, protocol, sdks/typescript)
cp .ai-agent/settings.example.json .ai-agent/settings.json   # Add your API configuration
npm run desktop:dev

When the desktop window opens, select Browser Automation and enter a prompt such as:

Open https://example.com and tell me the page title.

For sites that require authentication, see extension mode and pairing in the Browser Automation Guide.

Configuration quick reference

User-level configuration: ~/.ai-agent/settings.json. Project-level configuration: <workspace>/.ai-agent/settings.json, which takes precedence.

Common commands

Browser automation

Coding Agent can open web pages, click controls, fill forms, take screenshots, and read page content. By default, it launches an isolated browser with no configuration required. It can also control your own Chrome profile, allowing access to authenticated sites without login scripts.

npm start                 # Isolated mode works out of the box
npm run browser:pair      # Check the extension, to use your own Chrome profile

For complete instructions, see docs/readme/browser.md.

Production deployment

Deploy the Web UI with Docker. See deploy/README.md for details.

docker compose -f deploy/docker-compose.admin.yml --env-file deploy/.env up -d
# Open http://localhost:9999 (credentials: WEB_USERNAME / WEB_PASSWORD)

Project structure

├── start.js              # Unified entry point (HTTP / --acp / --stdio)
├── src/                  # Agent core (tools / core / services / server)
│   ├── tests/            #   Legacy test scripts, run through vitest (suites/, suites/live/)
│   └── dev/              #   Developer utilities (browser pairing, dev Chrome)
├── web/                  # React frontend (npm workspace)
├── desktop/              # Electron shell + electron-builder config
├── chrome-extension/     # Browser extension (extension mode)
├── protocol/             # Shared wire protocol (npm workspace)
├── sdks/                 # Client SDKs: typescript/ (npm workspace), python/
├── analytics/            # Telemetry service (Python / FastAPI)
├── integrations/         # Third-party bridges (stpl-lsp-bridge)
├── evals/                # SWE-bench / aider-polyglot evaluation harnesses
├── deploy/               # Docker images, compose files, workspace seeds
├── scripts/              # Build / staging scripts
├── docs/                 # VitePress documentation
└── .ai-agent/            # Project-level skills / agents / commands / settings.example.json

Documentation index

User guides

Development and architecture

License

Licensed under the MIT License. Copyright © 2026 Harry Liu.