Lite is a fast, local workspace for Claude Code, Codex on OpenAI, DeepSeek, or OpenRouter, Gemini CLI, Kimi Code, Qwen Code, and your shell. Keep agent sessions, files, and Git context together without repository indexing, telemetry, or a cloud service.
- Run Claude Code, Codex, Gemini CLI, Kimi Code, Qwen Code, and shell sessions side by side
- Run Codex against DeepSeek V4 Flash or Pro, or OpenRouter, without changing your default Codex provider
- Resume session tabs automatically after closing Lite or restarting your computer
- Undo a restart or close for eight seconds before Lite stops the terminal
- Authenticate once with each provider and reuse its existing local credentials
- Or save API keys for supported providers in Lite and skip their sign-in flows entirely
- Browse files on demand with language icons, then edit them with search, replace, and multi-cursor
- Preview rendered Markdown safely alongside source files
- See the active Git branch, worktree, and changed files
- Inspect per-session context and provider usage reported by Claude or Codex
- Install signed updates from inside Lite
- Switch between light and dark themes, terminal and code preview included
- Rebind any keyboard shortcut from Settings
Lite is intentionally quiet: idle means idle. It does not index your repository, watch every file, read provider tokens, or send telemetry.
Download Lite from the latest GitHub Release. Release assets are built from the same source for macOS, Windows, and Linux.
- Download the asset ending in
_darwin_aarch64.dmgfor an Apple silicon Mac. - Open the disk image and drag Lite into Applications.
- Open Lite from Applications. If macOS blocks the unsigned app, open System Settings → Privacy & Security, set Allow applications from to App Store & Known Developers, select Open Anyway for Lite, then confirm.
- Download the asset ending in
_windows_x64-setup.exe. - Run the installer, then open Lite from the Start menu.
- If Microsoft Defender SmartScreen appears for the unsigned app, select More info → Run anyway.
Download the portable asset ending in _linux_amd64.AppImage.
# AppImage
chmod +x Lite_*_linux_amd64.AppImage
./Lite_*_linux_amd64.AppImageAfter installing Lite, open Settings from the gear in the top bar and choose Check for updates under About to install signed updates and restart Lite. Lite never checks for updates in the background.
Install the provider CLIs you want to use:
- Claude Code
- Codex
- Gemini CLI
- Kimi Code — Windows also needs Git for Windows
- Qwen Code
Open each provider in Lite and complete its normal sign-in once. Every CLI keeps credentials in its own local store, so later Lite sessions reuse the same authentication. Lite never reads or copies those stores. The new-session dialog installs a missing CLI for you.
If you would rather use API keys, open Settings › API Keys and paste one for a supported provider. Lite keeps them in an owner-only file in its own data folder — the same shape Codex and Kimi already use for their credentials — and passes a key to a session through the environment variable that CLI already reads (ANTHROPIC_API_KEY, OPENAI_API_KEY, DEEPSEEK_API_KEY, OPENROUTER_API_KEY, GEMINI_API_KEY, or MOONSHOT_API_KEY). Qwen Code keeps ownership of its regional provider and authentication setup. Nothing is written into provider configuration, deleting a key takes effect on the next launch, and app updates keep the file since the updater replaces the bundle and not your data.
Codex · DeepSeek runs the Codex harness against DeepSeek V4 Flash or Pro through its OpenAI Responses API. Saving a DeepSeek key in Lite is enough — Lite then defines the provider for that launch only. To configure it yourself instead, use your own Codex configuration; a deepseek.config.toml profile in $CODEX_HOME is cleanest:
model = "deepseek-v4-flash" # Or "deepseek-v4-pro" for the most capable model
model_provider = "deepseek"
[model_providers.deepseek]
name = "deepseek"
base_url = "https://api.deepseek.com/"
wire_api = "responses"
experimental_bearer_token = "<your DeepSeek API key>"Leave preferred_auth_method and forced_login_method out. Codex applies those globally, and setting them signs you out of ChatGPT the next time it runs. Lite selects the DeepSeek provider per launch, so your default Codex provider and existing OpenAI sessions are untouched, and it never reads or stores the DeepSeek key.
Codex · OpenRouter works the same way: save an OpenRouter key in Lite or configure an openrouter Codex provider, and Lite launches OpenRouter's current OpenAI model route without changing your default Codex setup.
Choose a project folder, create a session, and work. Lite saves only the local metadata needed to restore tabs; provider conversation history remains owned by the provider CLI.
Install Bun, Rust, and the platform-specific Tauri prerequisites, then run:
git clone https://github.com/ultralytics/lite
cd lite
bun install
bun run tauri devUseful checks:
bun run check # Biome lint, native TypeScript check, and knip dead-code scan
bun run local # Separate Lite Dev app with a red icon and its own app data
bun run tauri build # Native installer for the current operating systemLite Dev follows origin/main directly instead of release assets. Its update button fetches main and,
when a newer commit exists, opens a visible shell session that fast-forwards and rebuilds the app.
The frontend uses React, shadcn/ui Nova with Base UI, Tailwind CSS, Biome, and tsgo. Tauri and Rust own local persistence, terminals, Git, files, and provider processes.
Bug reports and focused feature proposals are welcome in GitHub Issues. Please keep Lite's core rule in mind: the simplest complete solution wins.
Lite is available under the AGPL-3.0 License. For commercial licensing, contact Ultralytics Licensing.











