LMSMOB Chat (LM Studio Android App)

Android chat app for LM Studio. Connect your phone to an LM Studio server running on your PC and chat with local LLMs over your LAN. LMSMOB Chat is built for a local-first privacy workflow: your prompts and answers are processed by your LM Studio server on your own computer when you use local LM Studio models, with local Supertonic read-aloud and scheduled prompt Watch Jobs for recurring LM Studio checks.

Download the latest APK from the GitHub releases page.

Demo

Features

Local-first chat with LM Studio: inference runs on your PC, not a hosted chat service, when you use local LM Studio models.
Native Android app built with Kotlin and Jetpack Compose.
Works with LM Studio's OpenAI-compatible local server at /v1.
Supports LM Studio's native /api/v1/chat endpoint for server tools.
Connects to LM Studio MCP/plugin integrations such as mcp/local-web.
Tested companion LM Studio tools are published at mindylab/lm_studio_tools.
Displays MCP image outputs in chat, including web page screenshots, reverse image search results, and generated QR codes returned by LM Studio tools.
Refreshes loaded LM Studio models from the app settings.
Keeps chat sessions locally with folders, search, session switching, and reusable chat text presets for quickly appending common instructions.
Supports copy, edit, delete, and cancel actions in chat.
Adds optional transparent chat navigation buttons for moving between messages, one-time send-and-read actions, and controlling read-aloud from either side of the screen.
Can notify you when a response finishes while the app is in the background.
Imports and exports chat history as JSON.
Supports image attachments for vision-capable models.
Shares text, images, PDFs, DOC/DOCX, and XLS/XLSX files into the app from Android share sheets.
Converts supported documents into text, and can render PDFs as image attachments for vision-capable models.
Adds optional voice input, text-to-speech output, automatic answer read-aloud, one-time auto-read send, Supertonic TTS engine selection, chunked local playback controls with pause/resume, resume points, retry recovery, auto-scroll during read-aloud, and screen-awake playback for long assistant answers. For best performance, use the MindyLab optimized Supertonic Android engine, which includes service isolation and request-level processing controls used by LMSMOB Chat.
Provides opt-in phone assistant tools for URL opening, map routes, email/SMS drafts, phone dialing, calendar/reminder drafts, contact lookup, notification digest, local file search, and device status.
Adds alarm tools for reading the next Android alarm, opening the Clock alarm screen, and preparing alarm creation drafts.
Adds Watch Jobs: local notification monitors, scheduled notification-history checks, and scheduled LM Studio prompt checks with filters, AI classification, server-tool support, interval/daily/weekly/monthly/one-time scheduling, normal notifications, alarm-style full-screen alerts, once/today/no-end lifetimes, and confirmation before model-created jobs are saved.
Shows clickable source chips and clickable links in rich chat answers.
Exports Markdown tables from chat answers as XLSX files.
Appends a capability guide to prompts so the selected model knows which server tools, phone tools, vision, reasoning, voice, document, and date/time capabilities are currently enabled.
Shows context usage estimates and advanced generation settings including temperature, top-p, max tokens, context length, penalties, and seed.
Includes emulator and physical-device LAN setup paths.

Privacy

LMSMOB Chat does not need a cloud chat service for normal use. The app sends your messages to the LM Studio server URL that you configure, usually a PC on your own Wi-Fi/LAN, and LM Studio processes those prompts locally when you run local models. Chat history is stored on your Android device unless you export, share, or delete it.

Voice privacy is also improved when using Supertonic TTS: assistant answers can be spoken by the Supertonic Android TTS engine locally on your phone, instead of depending on a cloud text-to-speech provider. LMSMOB Chat can split long answers into local playback chunks, with quality and buffering controls for smoother read-aloud. The MindyLab optimized Supertonic Android engine is recommended for better long-answer processing. If you choose the Android system TTS engine, privacy depends on the TTS engine selected in Android settings.

MCP/plugin tools are opt-in and can change the privacy boundary. Local tools can stay on your PC, but web search, web fetch, YouTube transcript, maps, remote MCP, or other network tools may send the specific URLs, queries, or task data needed to external websites or services. Only enable integrations and allowed tools that you trust. Scheduled prompt Watch Jobs use the configured LM Studio server and may use those same server tools when a recurring task needs current web or tool data.

Roadmap

The 2026 roadmap now builds on the first phone-assistant tools shipped in v1.28, the Watch Jobs release in v1.40, scheduled prompt Watch Jobs in v1.60, and TTS send/recovery improvements in v1.61:

Web tools already work through compatible LM Studio MCP/plugin setup, with a tested local-web MCP server available in mindylab/lm_studio_tools. A packaged LMS/LM Studio web plugin is still planned to make setup easier.
Maps route drafts, contact lookup, notification digest, local file search, device status, and confirmable phone action drafts are available now.
Alarm tools and Watch Jobs are available now for opt-in notification monitors, scheduled LM Studio prompt checks, daily/weekly/monthly/one-time schedules, and stronger local alerts.
Phone location tools are still planned with explicit Android permission controls.
Broader cron-style task history, pause/resume controls, and safe follow-up phone actions are still planned.
A signed release APK workflow is planned for wider distribution.

See the full 2026 roadmap.

Quick Setup

Install the APK from the latest release.
Start LM Studio on your PC.
Load a model and start the local server from LM Studio.
Open LMSMOB Chat settings.
Set the LM Studio API URL and model.
If LM Studio has authentication enabled, add your LM Studio API token.

The Android emulator default URL is:

http://10.0.2.2:1234/v1

For a physical Android device, use your PC's LAN IP address:

http://YOUR_PC_LAN_IP:1234/v1

Example:

http://192.168.1.25:1234/v1

Server Tools And MCP

LM Studio 0.4.0+ can call MCP/plugin tools through its native /api/v1/chat endpoint. In LMSMOB Chat, open settings, enable Server tools, and add the LM Studio integration IDs you want to allow.

The tested LM Studio tools repo is mindylab/lm_studio_tools. It contains the local-web MCP server used with LMSMOB Chat, including web_search, web_fetch, web_search_and_fetch, youtube_transcript, qr_generate, qr_scan, reverse_image_search, and web_page_to_images, plus ready-to-copy mcp.json examples.

Server-side tools require an LM Studio API token. In LM Studio, create a permission token from Developer > Server Settings > Manage Tokens, then paste it into the app's API token field. The token must allow the MCP/tool permissions you want to use.

Privacy note: MCP/plugin integrations run outside the Android app, usually on the PC running LM Studio. Some tools are local, while others intentionally access the internet or remote services. Review each tool's behavior before enabling it, especially tools that search the web, fetch pages, call maps, or connect to remote MCP servers.

When Server tools is enabled, tap the refresh icon beside the model field. The app loads model IDs from /api/v1/models and prefers currently loaded LLM instances, which is what /api/v1/chat expects.

For an MCP server already configured in LM Studio's mcp.json, add one plugin ID per line:

mcp/local-web
mcp/gemma4-audio-python

Bare server labels are normalized by the app. For example, local-web becomes mcp/local-web.

For broad local tools, fill Allowed tools in the app settings with one tool name per line. The app will send:

{
  "type": "plugin",
  "id": "mcp/local-web",
  "allowed_tools": ["web_fetch"]
}

Against the tested server, mcp/local-web accepted these allowed tools:

web_search
web_fetch
web_search_and_fetch
youtube_transcript
qr_generate
qr_scan
reverse_image_search
web_page_to_images

For an ephemeral remote MCP, paste a JSON object or JSON array:

{
  "type": "ephemeral_mcp",
  "server_label": "huggingface",
  "server_url": "https://huggingface.co/mcp",
  "allowed_tools": ["model_search"]
}

When server tools are enabled, the app sends chat requests to:

http://HOST:1234/api/v1/chat

LM Studio server settings must allow the relevant MCP/plugin type.

Build

With Android Studio's JDK and SDK available:

.\gradlew.bat assembleDebug

The debug APK is generated under:

app/build/outputs/apk/debug/

To build a release APK:

.\gradlew.bat assembleRelease

The project can sign release builds when a local keystore.properties file is present. See docs/release-signing.md. Do not commit keystores, passwords, or keystore.properties.

Keywords

LM Studio Android app, LM Studio mobile client, Android local LLM chat, local AI Android app, OpenAI-compatible Android client, MCP Android chat, Kotlin Jetpack Compose AI chat, local server chat app.

License

Official LMSMOB Chat app builds are free for personal, non-commercial use. The source code remains proprietary to MindyLab MB. Redistribution, modification, commercialization, rebranding, company use, or source-code use requires prior written permission. See LICENSE.md.

Lmsmob Chat

README