Signal LM

A lightweight, local-first AI chat studio — runs in the browser or as a desktop app, connects to any LM Studio-compatible server, and extends onto Android with Vulkan GPU inference.

---

What is Signal LM?

Signal LM is an open-source front-end for local language models. It talks to an LM Studio server (or any OpenAI-compatible /v1 endpoint) and adds a layer of practical features on top of the raw API:

Feature	Description
Chat	Full conversation UI with streaming, code-block rendering, syntax highlighting, export, and persistent history
Folders / Workspace	Load a project folder into context — browse files, select what the model sees, and apply AI-generated edits with a review step before anything is written
Silent Context Helper	Pre-processes workspace files before each request, sending only the most relevant snippets instead of dumping the whole folder — reduces prompt size and speeds up inference
MCP (Model Context Protocol)	Add remote or local MCP tool integrations so the model can call external services during chat
Android + Vulkan Runtime	Run inference on-device on an Android phone via a native bridge; or use Hybrid mode to keep the PC server primary and fall back to / race against the phone
Telemetry panel	Right-side system sidebar shows live CPU, RAM, GPU, storage, and LM Studio model count from the local tel-server.js helper
Themes	System / Light / Dark, persisted in localStorage

---

Getting Started

Prerequisites

• Node.js (v18 or later)
• LM Studio running locally with the server enabled
  (default: http://localhost:1234/v1)

1 — Install dependencies

npm install

2 — Run as a desktop app (recommended)

npm run desktop

This opens Signal LM in an Electron window and starts tel-server.js automatically so the System Status panel can read CPU, RAM, GPU, and storage metrics without a separate terminal.

3 — Run in the browser (static / web)

Open index.html directly, or serve the folder with any static file server.
For the System Status panel to show hardware data, also start the telemetry helper:

node tel-server.js
# or double-click tel-server.bat

The helper listens on http://127.0.0.1:8766/status (local-only). The app works without it — the status cards just show an offline state.

---

Connecting to LM Studio

1. Open Settings → Connection
2. Set Base URL to your LM Studio server
   • Same machine: http://localhost:1234/v1
   • Phone or another device on the network: http://<server-ip>:1234/v1
3. Set API Key only if you enabled authentication in LM Studio (leave blank otherwise)
4. Click Save Connection → Test PC Connection + Load Models

---

Runtime Modes

Signal LM supports three inference backends, selectable from the Chat sidebar or Settings → Android Runtime:

Mode	Description
LM Studio server (default)	All requests go to the configured /v1 endpoint
Android Vulkan local only	All inference runs on the phone via the native bridge; no PC server needed
Hybrid boost	PC server is primary; the phone provides backup compute via one of the strategies below

Hybrid Strategies

Strategy	Behaviour
Off	PC server only (hybrid mode inactive)
Fallback	Try the PC first; if it fails or exceeds the timeout, run the same request on the phone
Race	Send the request to both at the same time; use whichever answers first and cancel the slower one
Task Splitting	PC generates the raw answer; phone generates a summary

Note: Hybrid mode does not merge PC VRAM and phone GPU memory into a single model. The phone runs a separate GGUF independently.

---

Folders & AI Edits Workflow

1. Go to the Folders page (#editor)
2. Click Select Folder to load a project directory
3. Check the files you want to give the model as context
4. Click Send Folder to Chat to inject them into the chat context
5. In chat, describe the edit you want — e.g. "Refactor the dark-mode CSS to use CSS variables"
6. The model returns full replacement file contents
7. Return to Folders → AI Folder Edits, review the diff, and click Apply Reviewed Edits
   — nothing is written to disk until you explicitly apply

The Silent Context Helper (Settings → Silent Helper) runs automatically before each request. It scores workspace files by relevance to your message and sends only compact, targeted snippets, keeping prompt size down and inference fast. Switch to Full selected files mode if you need the model to see an exact small file in its entirety.

---

MCP Integrations

1. Go to MCP (#mcp)
2. Enable MCP in chat
3. Add a Remote / Ephemeral MCP integration with a server URL (e.g. https://huggingface.co/mcp), or an LM Studio mcp.json Plugin using its plugin ID
4. Save and run MCP Smoke Test to verify
5. Chat now routes through LM Studio's native REST endpoint with tool definitions injected

---

System Status Panel

The right-side System Status sidebar shows:

• CPU model, logical cores, and usage %
• System RAM used / free / total
• GPU usage and VRAM (via nvidia-smi or Windows GPU engine counters)
• Storage used / free / total for the project drive
• LM Studio reachability and loaded model count

The desktop app starts tel-server.js automatically. For the browser version, run it manually or via tel-server.bat. The helper binds only to 127.0.0.1 and never sends data off your machine.

Environment overrides for tel-server.js:

Variable	Default	Description
SIGNAL_LM_TELEMETRY_PORT	8766	Port the helper listens on
SIGNAL_LM_API_BASE_URL	http://localhost:1234/v1	LM Studio endpoint to probe
SIGNAL_LM_API_KEY	(empty)	API key if LM Studio auth is enabled

---

Building a Windows Installer

npm run dist

The installer is written to dist_new\Signal-LM-Setup-v1.0.5.9.exe.

---

Publishing Desktop Releases

Desktop installers are published as GitHub Release assets — do not commit them to the repo.

npm version patch        # bumps version in package.json
git push origin main --tags

Pushing a v* tag triggers .github/workflows/desktop-release.yml, which builds the Windows installer on GitHub Actions and uploads it to the release.

Latest stable installer URL (used by the app header and Settings page):

https://github.com/falabellamichael/Signal-LM/releases/latest/download/Signal-LM-Setup.exe

To update an installed desktop app, download the latest installer and run it — it replaces the existing installation in-place.

---

Android Native Bridge (Developer Reference)

To connect an Android WebView app as an inference runtime, expose one of these JavaScript objects before index.html loads:

window.lmStudioLiteNative
window.NativeInferenceBridge
window.AndroidInferenceBridge

Required methods

getHardwareStatus(): string        // JSON: GPU name, RAM, thermal state
listModels(): string               // JSON: array of available model paths
chatCompletion(payloadJson: string): string
generate(payloadJson: string): string   // optional fallback
cancelGeneration(): void               // recommended for race mode

File/workspace methods (optional but needed for Folders)

selectFolder(): string
readFile(path: string): string
writeFile(path: string, content: string): void
writeFiles(payloadJson: string): void

See ANDROID_VULKAN_RUNTIME.md for the full payload schema and runtime tuning starting points.

---

Project Structure

File / Folder	Purpose
index.html	Single-page app shell (Chat, Folders, MCP, Settings views)
index.js	Core chat logic, model loading, workspace, attachments
index.css	Main stylesheet
signal-lm-commands.js	Slash-command system for chat
signal-lm-mcp-*.js	MCP pipeline, auth, chat bridge, file-path resolution
editor.js / editor.css	Folders page logic and styles
mcp.js / mcp.css	MCP configuration page
settings.js / settings.css	Settings page
system-sidebar.js / .css	System Status panel
shared.js / shared.css	Utilities and tokens shared across pages
tel-server.js	Local telemetry server (CPU, RAM, GPU, storage)
desktop-main.js	Electron main process
SignalLM-Android/	Android companion app source
ANDROID_VULKAN_RUNTIME.md	Android bridge API reference
SILENT_CONTEXT_HELPER.md	Context helper internals
TELEMETRY_HELPER.md	Telemetry server reference

---

License

GPL-3.0