A clean, cross-platform desktop chat client for LM Studio's local REST server.
Chat with any locally running language model — with streaming, file attachments, MCP integrations, and full sampling control.
- Four API modes — Native LM Studio
/api/v1/chat, OpenAI/v1/chat/completions, OpenAI/v1/responses, and Anthropic/v1/messages(experimental) - Streaming responses — live token streaming with stop and status indicators
- File & image attachments — drag & drop, paste, or file-picker; text files are inlined into the prompt; image files are sent as vision input
- MCP integrations — add ephemeral or plugin-based Model Context Protocol integrations directly from Settings
- Model manager — load, unload, and download models by ID; configure context length, eval batch size, KV-cache placement, flash attention and more from the UI
- Full sampling control — Temperature, Top-P, Top-K, Repeat Penalty, Frequency Penalty, Presence Penalty, Max Tokens, Context Length, Reasoning Effort
- Stateful chat — preserve server-side conversation state across turns (native +
/v1/responsesmodes) - System prompt — configurable per session
- Raw JSON override — merge arbitrary fields into the request payload for advanced use
- Export / Import — back up and restore all chats, messages and settings as a single JSON file
- Light / dark / system theme — with native Windows title-bar overlay color sync
- Localization — English and Russian interface languages, easily extensible
- Update checker — checks for new releases and shows a non-intrusive prompt
- Onboarding wizard — guided first-run setup: connect to server, pick a model, choose language
- Responsive layout — three-column desktop grid collapses to off-canvas drawers on tablets; dedicated mobile layout for phones (including Capacitor/Android target)
- Secure by default —
contextIsolation: true,nodeIntegration: false; API tokens are kept in memory only unless the user explicitly opts in to localStorage persistence
| Platform | Status |
|---|---|
| Windows (10/11) | ✅ Fully supported — NSIS installer, native title-bar overlay |
| macOS | ✅ Supported — DMG, hiddenInset title bar, traffic-light buttons; must build on a Mac |
| Linux | ✅ Supported — AppImage; must build on Linux |
| Android (Capacitor) | |
| Browser (web) |
- LM Studio running with its local server enabled (default:
http://localhost:1234) - Windows 10 or later, macOS 11+, or a modern Linux distribution (depending on the package)
Go to Releases and download the package for your platform:
| Platform | File |
|---|---|
| Windows | LM Client Setup x.x.x.exe (NSIS installer) |
| macOS | LM Client-x.x.x.dmg |
| Linux | LM Client-x.x.x.AppImage |
Windows: Run the installer — it creates a desktop shortcut and a Start Menu entry.
macOS: Open the DMG and drag the app to /Applications.
Linux: Make the AppImage executable (chmod +x) and run it.
See Build Instructions below.
# 1. Clone the repository
git clone https://github.com/<your-username>/lm-client.git
cd lm-client
# 2. Install dependencies
npm install
# 3. Start in development mode (opens Electron window)
npm startNote: The app loads
www/index.htmldirectly as a local file, so no local dev server is required.
The project uses electron-builder to produce distributable packages.
# Install dependencies (if not done already)
npm install
# Windows — NSIS installer (run on Windows)
npm run dist
# Windows — unpacked directory, faster for testing
npm run pack
# macOS — DMG (must run on macOS)
npm run dist:mac
# Linux — AppImage (must run on Linux)
npm run dist:linuxBuild output is placed in the dist/ directory.
Cross-compilation is not supported by electron-builder for most targets. Build each platform on its native OS.
Code signing (Windows): Set the
CSC_LINKandCSC_KEY_PASSWORDenvironment variables before building to sign the installer.
Code signing (macOS): SetCSC_LINK,CSC_KEY_PASSWORDand optionallyAPPLE_ID/APPLE_APP_SPECIFIC_PASSWORDfor notarization.
The constant COMPILE_AS near the top of www/app.js controls the window-chrome behaviour and some layout decisions at runtime:
const COMPILE_AS = 'WINDOWS'; // change before building for another platform| Value | Use case |
|---|---|
WINDOWS |
Electron on Windows — hidden title bar + native overlay (min/max/close) |
MACOS |
Electron on macOS — hidden-inset title bar, traffic-light buttons |
LINUX |
Electron on Linux — no window-chrome adjustments |
AUTO |
Runtime detection (useful for browser / Capacitor builds) |
android_apk |
Capacitor Android build, forces mobile layout |
Before building for a non-Windows platform, update COMPILE_AS to the appropriate value. The setting affects drag regions, reserved space for window controls, and the title-bar overlay API calls.
- Start LM Studio and enable the local server (Developer tab → Start server).
- Open LM Client.
- On first run, the onboarding wizard will guide you through connecting to the server, picking a model, and selecting the interface language.
- Select your API mode from the top bar (Native, OpenAI, or Anthropic).
- Select a model from the model dropdown.
- Type your message in the composer and press Enter (or Shift+Enter for a new line).
- To attach a file or image, click the paperclip button or drag & drop a file onto the window.
- Adjust generation parameters in the Settings panel on the right.
- Click New chat in the sidebar to start a fresh conversation.
- Right-click (or long-press on mobile) a chat to rename, pin, or delete it.
- Use the search bar at the top of the sidebar to filter chats.
Click the Export/Import button in the top bar to save all chats and settings to a JSON file, or to restore from a previously exported file.
All settings are accessible from the Settings panel (gear icon, top-right).
| Setting | Description |
|---|---|
| Base URL | Address of the LM Studio server (default: http://localhost:1234) |
| API token | Optional Bearer token if the server requires authentication |
| Persist token | Save the token to localStorage (marked as a risk — see Security Notes) |
| API mode | Switch between Native, OpenAI Chat, OpenAI Responses, and Anthropic endpoints |
| Temperature / Top-P / Top-K | Standard sampling parameters |
| Repeat Penalty | Penalise repeated tokens (native mode) |
| Frequency / Presence Penalty | OpenAI-compatible penalty parameters |
| Max output tokens | Cap on generated tokens (-1 = no limit) |
| Context length | Override the model context window (native mode) |
| Reasoning effort | low / medium / high / on — passed when the model supports it |
| Stream responses | Toggle streaming on/off |
| Stateful chat | Maintain server-side conversation state (native + responses modes) |
| System prompt | Prepended system message for every conversation |
| MCP integrations | Add Model Context Protocol plugins or ephemeral MCP servers |
| Raw JSON override | Arbitrary JSON merged into every request payload |
| Language | Interface language: English / Russian |
| Theme | Light / Dark / System |
Settings are persisted to localStorage (key: lmstudio-client:settings).
Chat history is stored in IndexedDB (database: lmstudio-client), with a localStorage fallback.
lm-client/
├── assets/
│ └── icon.icns # macOS / Linux application icon
├── build/
│ └── icon.ico # Windows application icon
├── screenshots/
│ ├── main-dark.png
│ └── main-light.png
├── www/
│ ├── lang/
│ │ ├── en.js # English i18n bundle (pre-bundled JS)
│ │ ├── en.json # English i18n source
│ │ ├── ru.js # Russian i18n bundle
│ │ └── ru.json # Russian i18n source
│ ├── app.js # All application logic (single-file)
│ ├── index.html # UI markup
│ └── styles.css # All styles and design tokens
├── main.js # Electron main process (platform-aware)
├── preload.js # Electron preload script (contextBridge, Windows only)
└── package.json
- Context isolation is enabled (
contextIsolation: true) and Node.js integration is disabled (nodeIntegration: false) on all platforms. On macOS,sandbox: trueis also enabled. The preload script (contextBridge) is loaded only on Windows, where it exposes a singlewindow.electronTheme.setTheme()call to sync the native title-bar overlay colour with the app theme. On macOS and Linux no preload bridge is needed. - External links opened from the renderer are routed to the system browser via
shell.openExternal(); the app itself never navigates to external URLs. - API token storage: By default the API token is kept in memory only and is never written to disk. The user can opt in to
localStoragepersistence via "Persist token to localStorage" in Settings. This option is clearly labelled as a risk in the UI, becauselocalStorageis readable by anyone with access to the browser profile or OS user account. - No remote code execution: The app loads only local files (
www/index.html). It makes outbound HTTP requests solely to the configured LM Studio server and tolmclient.app/api/v1/for optional version checks and anonymous telemetry pings (platform + version + language + theme, no PII). - Telemetry: On launch and session end, the app sends an anonymous ping to
lmclient.app/api/v1/stats.phpcontaining only: event name, platform, version, interface language, theme, and bucketed session duration/message count. No IP address, model name, chat content, or API keys are ever transmitted. You can review theStatsobject inwww/app.jsto verify.
This roadmap reflects planned or possible directions; nothing here is a commitment.
- macOS and Linux packaged releases
- Keyboard shortcuts (new chat, toggle sidebar, stop generation)
- Chat search across message content
- Markdown export of individual chats
- Additional interface languages
- Conversation branching / message editing with history
- Custom themes / accent color picker
- Android (Capacitor) public release
Contributions are welcome. Please read CONTRIBUTING.md before opening a pull request.
Quick start:
- Fork the repository.
- Create a branch:
git checkout -b fix/your-descriptionorfeat/your-description. - Make your changes. Keep the scope focused — one logical change per PR.
- Test manually on your platform before submitting.
- Open a pull request against
mainwith a clear description of what changed and why.
Bug reports and feature requests: Use GitHub Issues.
This project is licensed under the MIT License — see the LICENSE file for details.
MIT License
Copyright (c) 2026 ntirl
LM Client is an independent, unofficial client application. It is not affiliated with, endorsed by, or officially connected to LM Studio or Anthropic. All product names, trademarks, and registered trademarks are the property of their respective owners.
This software communicates exclusively with servers you configure. The developer is not responsible for content generated by language models, for any data sent to or received from the configured server, or for any consequences of using this software.