lmstudio-firefox-proxy

A lightweight Rust proxy that bridges Firefox's AI sidebar to a local LM Studio instance.

Firefox's AI chatbot sidebar sends GET /?q=<prompt> requests to the configured provider URL. LM Studio expects OpenAI-compatible POST /v1/chat/completions requests. This proxy translates between the two — streaming tokens in real time with a polished, rendered UI.

🪄 100% Vibe-coded

This project was 100% vibe-coded — every single line of Rust, HTML, CSS, and JavaScript was written by AI through conversational prompting. No code was written by hand.

Built with:

• GitHub Copilot (agent mode)
• Claude Opus 4.6 by Anthropic

The human provided the idea, direction, and feedback. The AI wrote the code.

Features

• Streaming — Responses appear token-by-token as the model generates them (SSE)
• Rendered Markdown — Code blocks with syntax highlighting, tables, lists, and more
• Quick follow-up questions — After each response, ask another question in-page (for example "summarize this website") using the original input and recent turns as context
• Thinking model support — Reasoning is shown in a collapsible panel during generation, then auto-collapsed once reasoning completes. Supports both the reasoning_content field (LM Studio 0.4.8+) and <think> tags in older versions
• Dark / Light mode — Follows the system prefers-color-scheme automatically
• Fully offline — All frontend dependencies (marked.js, highlight.js) are vendored and embedded in the binary
• Single binary — No runtime dependencies, no config files, just run it

demo.mp4

Installation

Pre-built binaries

Download the latest release for your platform from the Releases page:

Platform	File
Linux x86_64	lmstudio-firefox-proxy-x86_64-unknown-linux-gnu.tar.gz
Linux aarch64	lmstudio-firefox-proxy-aarch64-unknown-linux-gnu.tar.gz
macOS x86_64	lmstudio-firefox-proxy-x86_64-apple-darwin.tar.gz
macOS Apple Silicon	lmstudio-firefox-proxy-aarch64-apple-darwin.tar.gz
Windows x86_64	lmstudio-firefox-proxy-x86_64-pc-windows-msvc.zip

Build from source

Requires Rust 1.85+.

cargo install --git https://github.com/blu3r4y/lmstudio-firefox-proxy

Or clone and build:

git clone https://github.com/blu3r4y/lmstudio-firefox-proxy
cd lmstudio-firefox-proxy
cargo build --release

The binary will be at target/release/lmstudio-firefox-proxy (.exe on Windows).

Usage

# Use defaults (listen on 127.0.0.1:8000, LM Studio at localhost:1234)
lmstudio-firefox-proxy

# Specify a model explicitly
lmstudio-firefox-proxy --model "lmstudio-community/gemma-3-27B-it-qat-GGUF"

# Custom listen address and LM Studio URL
lmstudio-firefox-proxy --listen 127.0.0.1:9090 --lmstudio-url http://192.168.1.100:1234

All options can also be set via environment variables:

Flag	Env var	Default
--listen / -l	LISTEN_ADDR	127.0.0.1:8000
--lmstudio-url	LMSTUDIO_URL	http://localhost:1234
--model / -m	MODEL	(empty — uses LM Studio's loaded model)

Running at Startup (Windows)

To have the proxy start automatically when you log in, create a scheduled task using Task Scheduler. Replace the paths and arguments as needed for your setup.

Program:

C:\Program Files\PowerShell\7\pwsh.exe

Arguments:

-WindowStyle Hidden -ExecutionPolicy Bypass -Command "Start-Process -FilePath 'C:\PATH\TO\lmstudio-firefox-proxy.exe' -ArgumentList '--model qwen/qwen3.5-9b' -WindowStyle Hidden"

Compatibility

Tested with LM Studio 0.4.8 (latest). Older versions that use the same OpenAI-compatible API should also work — thinking models are supported regardless of whether the server sends reasoning via the reasoning_content field (new) or <think> tags (old).

Firefox Configuration

1. Open about:config in Firefox
2. Set browser.ml.chat.enabled to true
3. Set browser.ml.chat.hideLocalhost to false
4. Set browser.ml.chat.provider to http://127.0.0.1:8000 (or whichever address the proxy listens on)
5. Open the AI chatbot sidebar (Ctrl+Alt+X or via the sidebar menu)

You should see a "Proxy is running" landing page. Select text on any page and use the "Ask AI" context menu, or use the sidebar directly.