Rakh is a desktop AI coding agent for local codebases. It combines a React/Vite frontend with a Tauri/Rust backend to give you a multi-tab agent workspace with tool approvals, git worktree isolation, an integrated terminal, durable artifacts, and specialized subagents.

• Multiple independent agent tabs, each with isolated state and chat history
• OpenAI, Anthropic, and OpenAI-compatible provider support
• Safe file edits and shell commands through explicit approval gates
• Global MCP server registry in Settings, with per-run MCP tool discovery
• Automatic git worktree setup before agent-driven code changes
• Built-in subagents for planning, review, security, copy, and GitHub tasks
• Durable artifacts for plans, reports, and other structured outputs
• Structured JSONL logging with per-run traces and backend persistence
• Integrated terminal, voice input, theme system, and session restore

• React 19 + Vite
• Tauri 2 + Rust
• Vercel AI SDK
• Jotai
• xterm.js
• Tailwind CSS 4

• Node.js 20+
• npm
• Rust toolchain
• Tauri system prerequisites for your platform

Install dependencies:

npm install

Run the web UI only:

npm run dev

Run the desktop app in development:

npm run tauri:dev

Structured logs are written by the desktop runtime to:

• release: ~/.rakh/logs/rakh.log
• debug: ~/.rakh-dev/logs/rakh.log

Example:

tail -f ~/.rakh/logs/rakh.log | grep '"tool-calls"'

Rakh does not call model APIs until you configure a provider.

You can:

• add an OpenAI, Anthropic, or OpenAI-compatible provider in Settings
• import OPENAI_API_KEY or ANTHROPIC_API_KEY from your environment if they are already set

Rakh can also discover tools from globally configured MCP servers.

In Settings -> MCP Servers you can:

• add stdio or streamable HTTP MCP server configs as JSON
• test discovery before saving
• enable Save returned files as artifacts if you want MCP-returned images or files to be stored in the artifact pane instead of being passed back into model context raw

The model picker is built from the static catalog in src/agent/models.catalog.json plus any cached model list for OpenAI-compatible providers.

1. Launch the app and open Settings if no provider is configured.
2. Add or import a provider.
3. Create a new session, choose a project folder, and pick a model.
4. Start chatting with the agent.
5. Approve edits, commands, or worktree creation when prompted.

Useful built-in slash commands:

• /plan
• /review
• /security
• /copywrite
• /github

Install the launcher from Settings -> Updates -> Command-line launcher.

Once installed:

• rakh opens or focuses the desktop app
• rakh <path> opens the app at that directory; if you pass a file, Rakh uses its parent directory
• rakh -a <path> opens the app at that directory and also adds it to your saved projects

The launcher lives under ~/.rakh/bin or ~/.rakh-dev/bin. Install also tries to add that directory to your PATH automatically. If your terminal is already open, restart it before using rakh.

Common commands:

npm run dev
npm run build
npm run lint
npm run typecheck
npm run test
npm run test:all
npm run tauri:dev
npm run tauri:build

Run Rust tests directly:

cd src-tauri && cargo test

• docs/artifacts.md - durable artifact model and validation flow
• docs/logging.md - structured log schema, storage, query APIs, and CLI usage
• docs/macos-release-signing.md - Apple signing and notarization credentials for macOS releases
• docs/subagents.md - subagent registry, contracts, and execution model
• docs/tauri-updater-release.md - updater signing, GitHub secrets, and rollout checks
• docs/windows-release-signing.md - Windows code-signing options and credentials for releases
• src/DESIGN_SYSTEM.md - UI primitives and token rules
• src/THEMING.md - theme/token implementation notes

• src/ - React UI, agent runtime, tools, and styles
• src-tauri/src/ - Rust commands and desktop integration
• docs/ - artifact and subagent documentation