Ask LM Studio — Local AI Sidebar for Firefox

Use LM Studio local models directly in the Firefox sidebar. Select Ask LM Studio from the sidebar controls and chat privately, offline, with no cloud.

This fills the gap in Firefox's built-in AI chatbot sidebar which only ships a few cloud providers (ChatGPT, Claude, Gemini, Copilot, Le Chat Mistral). Firefox WebExtensions cannot inject into that built-in provider dropdown, so this extension provides the standard WebExtensions way: a sidebar_action that lives in the same sidebar switcher.

What you get

• 📎 Sidebar entry “Ask LM Studio” — View → Sidebar → Ask LM Studio, or click the sidebar launcher → Ask LM Studio, or toolbar button
• 💬 Full chat UI talking to LM Studio's OpenAI-compatible server (/v1/models, /v1/chat/completions with streaming)
• 🔄 Model picker auto-loaded from LM Studio
• 📄 “Attach page” in sidebar + right-click menus (on by default — if you install the enterprise policy, uncheck “Show Ask LM Studio in the right-click menu” in Options so you don't get a duplicate menu)
• ⚙️ Options page for Base URL, model, system prompt, temperature, thinking mode
• 🔒 Local-first — defaults to http://localhost:1234, no telemetry. If you set a custom Base URL (e.g. LAN http://192.168.x.x:1234 or remote), page selections/context will go there — keep it local for full privacy, and use an https:// Base URL if the server isn't on your machine so traffic is encrypted in transit (see Privacy). Note: the extension requests http://*/* + https://*/* host access (shown as one “Access your data for all websites” item) so 📄 Attach page can run on any page and so chat can reach LM Studio — the local server sends no CORS headers, so without host access the browser would block it. Chat still goes only to your configured Base URL.

1. LM Studio setup (5 min)

1. Install LM Studio from https://lmstudio.ai, download a model (e.g. Llama 3.2, Mistral, Qwen).
2. Go to Developer tab → Start Server, port 1234.
   • Enable Serve on local network if needed, and Enable CORS (required for fetch from Firefox).
3. Load your model (or lms load on CLI).
4. Verify: curl http://localhost:1234/v1/models should list your model.

2. Install extension in Firefox

Temporary (dev)

1. about:debugging → This Firefox → Load Temporary Add-on → manifest.json
2. Open sidebar launcher (View → Sidebar) → Ask LM Studio. Or click toolbar button.
3. Status bar shows ✓ Connected (if not, ··· → Reload LM Studio).

Permanent

• web-ext build → upload .zip to about:addons or sign via AMO, or use ESR policies.

3. Use it

• Select model from dropdown, type, Enter to send.
• Quick prompts — Summarize / Explain this / Quiz me / Proofread, same labels and prompt texts as the built-in chatbot (Summarize uses typed composer text first, otherwise the attached page).
• Attach page (footer) → asks about current tab.
• ··· menu: Reload LM Studio, New chat, Show shortcut when selecting text, Settings…
• Select text on any page → right-click → Ask LM Studio.

4. Show LM Studio in Firefox's native AI chatbot (alongside Gemini / ChatGPT)

No WebExtension — including this one — can inject into the built-in provider dropdown. That list is controlled only by Firefox prefs / Enterprise Policy. This repo now ships the official workaround:

• policies/lm-studio.json — ready-made AIChatbot policy that adds “LM Studio (local)” next to the built-ins. No hand-copy needed:
  • Linux: sudo bash scripts/install-policy-linux.sh
  • Windows (admin PowerShell): scripts\install-policy-windows.ps1
  • macOS: bash scripts/print-policy-macos.sh (prints MDM payload — macOS needs MDM/managed profile, no silent local install)
  • Requires Firefox 149+. Verify in about:policies.
• Manual about:config test (no policy file):
  • browser.ml.chat.enabled = true
  • browser.ml.chat.sidebar = true
  • browser.ml.chat.hideLocalhost = false
  • browser.ml.chat.provider = http://localhost:3000/?temporary-chat=true
  • The options page shows this extension's own moz-extension://.../sidebar/sidebar.html URL too — you can paste that as browser.ml.chat.provider to iframe this UI inside the native chatbot for a quick test. Experimental: Firefox may refuse to iframe moz-extension:// pages from the chatbot (no web_accessible_resources is declared), so confirm it loads in your version before relying on it.

Important: the native chatbot just iframes a web-chat URL, not the raw LM Studio API. Since :1234 is API-only:

docker run -p 3000:8080 ghcr.io/open-webui/open-webui:main
# Open WebUI → Settings → Connections → OpenAI → http://host.docker.internal:1234/v1

Then the policy URL http://localhost:3000/?temporary-chat=true gives you LM Studio models inside the native dropdown. Full guide: https://docs.openwebui.com/tutorials/integrations/dev-tools/firefox-sidebar

This extension remains the zero-dependency way — no Docker / Open WebUI needed, talks straight to localhost:1234 from its own sidebar entry.

Troubleshooting

• Can't reach LM Studio: Server not started? Wrong port? Check options Base URL ends with /v1.
• CORS / empty response: In LM Studio Server settings enable CORS, restart server.
• No model: Load model in LM Studio first, then ··· → Reload LM Studio.
• Sidebar not showing: Firefox 140+ required (the built-in data-collection consent the manifest uses needs it). View → Sidebar → Ask LM Studio. browser.ml.chat prefs are for native chatbot only, not needed for this extension.

Files

manifest.json
sidebar/sidebar.html|css|js
background/background.js
options/options.html|js|css
shared/config.js
shared/base.css
shared/firefox-tokens.css
icons/

The sidebar and options page are styled with Firefox's own design tokens: shared/firefox-tokens.css is the chrome design-system (tokens-shared.css + tokens-brand.css) bundled from Firefox 157, because extension pages can't load chrome://global/skin/…. Both pages load shared/base.css, which sets color-scheme: light dark, and use light-dark() tokens, so they follow Firefox's light/dark theme.

Privacy (data collection disclosure)

Ask LM Studio has no telemetry, analytics, accounts, or third-party services. The manifest's data_collection_permissions declares exactly what can leave the browser:

• Why: solely to generate the model's reply — the extension's whole purpose. Nothing else (browsing history, identifiers, device info, usage statistics) is ever sent anywhere.
• Storage: settings and chat history stay in the browser; the queued right-click prompt uses in-memory (storage.session) data that Firefox clears on exit, so nothing from private windows is retained across sessions. Whatever your configured server keeps is under your control.
• Control: keep the default localhost Base URL and nothing leaves your machine. For a remote server, use an https:// Base URL so data is encrypted in transit.
• Sharing: none. There are no third parties; the destination server is one you chose.

License: GPL-2.0 — local-first. By default no data leaves your machine; a custom Base URL sends prompts there. Exception: shared/firefox-tokens.css is bundled Mozilla code under MPL-2.0.