Key takeaways
- You can run an AI model entirely on your own machine and use it inside Shift, no API subscription required.
- LM Studio turns your computer into a local AI server. Shift, or any Chromium-based extension, can talk to it.
- The core requirement is enabling CORS in LM Studio so a browser can actually reach the local server.
- You'll need a Chrome-compatible, OpenAI-format extension to bridge the connection since this isn't part of Shift's built-in AI.
- Local models trade some speed and quality for full data privacy and zero ongoing cost.
- A decent GPU with 16GB+ VRAM (or a modern Apple Silicon Mac) makes the experience genuinely usable.
- Most connection issues come down to a missing CORS setting or a mismatched model name.
If you've ever wanted AI in your browser without your prompts leaving your laptop, this is the setup for you. It's a bit more hands-on than flipping a toggle, but once it's running, you get a private, offline-capable AI assistant that costs nothing per query.
Why bother running AI locally
A few real reasons people go this route instead of a hosted model:
- Privacy. Nothing you type leaves your machine.
- Cost. No subscription, no per-token billing.
- Offline access. Once a model is downloaded, you don't need a connection to use it.
The tradeoff: local models need decent hardware, and won't match the biggest cloud models on raw capability. For everyday tasks, summarizing, drafting, quick Q&A, that gap rarely matters.
How to keep your search engine private
Mainstream engines track your every move, but you can reclaim your data. Use private search tools and structured browsers to isolate your digital identity and stop surveillance.
Read MoreWhat you'll need before you start
A few things to have in place first:
- Shift Browser, updated to the latest desktop version. Since Shift supports standard Chrome extensions, it can run the same AI extensions you'd use in any Chromium browser.
- LM Studio, downloaded for your OS (Mac, Windows, or Linux).
- Enough hardware to run a model comfortably:
- 16GB of system RAM is a reasonable minimum for smaller models.
- A GPU with 8GB+ VRAM speeds things up. 16GB or more gives headroom for larger models.
- Apple Silicon Macs (M1 and later) handle local models well thanks to unified memory.
If your machine is on the lighter end, stick to smaller quantized models. They'll run faster and still handle most day-to-day tasks fine.
Setting up LM Studio
Once LM Studio is installed, there are three steps: get a model, load it, and turn on the local server.
Download a model Open the search tab inside LM Studio and pull down a GGUF model. Llama 3, Qwen, and Mistral are all solid, well-supported starting points. If you're not sure which to pick, a mid-sized quantized version (like a Q4) is usually the best balance of speed and quality for local use.
Load the model Select your downloaded model and load it into memory. LM Studio will use your GPU if it detects one, otherwise it falls back to CPU.
Turn on the local server This is the part that actually makes the browser connection possible:
- Go to the Developer or Local Server tab in LM Studio.
- Enable CORS (Cross-Origin Resource Sharing). This step matters more than it sounds. Without it, your browser will block requests to the local server by default.
- Start the server. By default, it runs at http://localhost:1234/v1.
Leave LM Studio running in the background. It needs to stay open for Shift to reach it.
Recent versions of LM Studio also include a web UI for managing models and the server, separate from the desktop app. Either works, but the desktop app is the more reliable place to toggle CORS and watch logs while testing. If you're new to LM Studio, spend a few minutes in the app's hub first, it's where you browse models, compare file sizes, and see quantization options before committing to a download.
Configuring Shift to talk to your local model
This part happens outside Shift AI itself. Shift's built-in AI is designed to work out of the box with its own infrastructure and privacy layer, so pointing it at a local server isn't part of that feature. Instead, this workflow uses Shift's support for standard Chrome extensions.
Install an OpenAI-compatible AI extension, something like Sider or UseChatGPT.AI, from the Chrome Web Store. Since Shift is Chromium-based, these install and run the same way they would in any other Chromium browser. Once installed, open the extension's settings from within Shift. This is where you'll point it at your own machine instead of a hosted API.
Shift AI is officially live: Built on Cloudflare's privacy-first, secure, and scalable infrastructure
Shift AI is now live, built on Cloudflare's Privacy Proxy to fully anonymize user data, bridging the critical AI trust gap.
Read MoreBridging the connection
Inside the extension's settings, there are usually three fields to update:
- API endpoint: Replace the default OpenAI URL with your LM Studio address, typically http://localhost:1234/v1.
- API key: Enter a placeholder value like lm-studio. LM Studio doesn't check for real authentication on local requests, but most extensions require something in that field.
- Model name: Type in the exact name of the model you loaded in LM Studio, or select it from the list if the extension can auto-detect local models.
Save the settings, and the extension should now be routing requests to your machine instead of the cloud.
A note on LM Studio plugins
LM Studio also supports plugins that extend what your local model can do, like giving it access to local files or basic tool calls. You don't need any of that for this setup. The plugins list is worth exploring later if you want your local model doing more than answering questions in the browser, but it's a separate track from getting Shift connected. Keep the first pass simple: model loaded, server running, CORS on.
Testing and troubleshooting
Run a test prompt. Highlight a block of text on any page in Shift and use the extension to summarize or explain it. A response back means the connection works end to end.
Watch the LM Studio logs. Keep the Local Server tab open while testing. You should see the incoming request logged in real time. If nothing shows up, the request isn't reaching LM Studio, usually pointing back to the endpoint URL or CORS setting.
Common issues and fixes
| Problem | Likely cause | Fix |
| Connection timeout | CORS not enabled | Re-check the Local Server settings in LM Studio |
| "Model not found" error | Model name mismatch | Copy the exact name shown in LM Studio |
| Slow or cut-off responses | Context length too short | Increase the context window in LM Studio's model settings |
| Extension shows no response | Server not running | Confirm LM Studio's server is started, not just the model loaded |
Most setups work fine once CORS is on and the model name matches exactly. If you're still stuck, restarting both LM Studio and the browser tab clears up a surprising number of one-off issues.
Going beyond localhost
Everything above assumes Shift and LM Studio run on the same computer. A few common follow-ups once people get that far:
- Connecting on your local network: if LM Studio runs on a different machine, say a desktop with a strong GPU, while you browse on a laptop, point the extension at that machine's local IP instead of localhost, as long as the port is reachable.
- Connecting to LM Studio remotely: technically possible, but exposing a local server to the open internet isn't something LM Studio was built for. If you need remote access, put it behind a VPN or an authenticated reverse proxy rather than opening the raw endpoint.
- Web search: LM Studio itself doesn't fetch live results. If your extension supports web search, it runs that search separately and feeds the results to your local model as context, worth knowing so you're troubleshooting the right layer if results look off.
Where this fits into your setup
Running a local model isn't a replacement for Shift AI, it's a different tool for a different job. Shift AI is built into the browser for fast, context-aware help with no setup: it already knows the page you're on and keeps your data private by default.
This LM Studio route is for people who want to go further, running their own model on their own hardware for specific privacy or offline needs. If that level of control appeals to you, it's worth remembering that's what Shift is built for elsewhere too. Spaces let you separate this kind of technical setup from your regular work, and the same drag-and-drop control that shapes your layout applies to how you organize the tools you bring in, local AI included.
