Lantern

Lantern™

Your own AI, running on your computer via Ollama.

Download for Windows
This won't work on Mac, Linux, phones, or tablets — pick the right tab above for your computer.
Free Runs via Ollama No API key Windows

Getting started

  1. Download and unzip the file above, anywhere you like.
  2. Install Ollama — free, from ollama.com/download. This is the free program that actually runs the AI.
  3. Open the unzipped folder and double-click Start Lantern.bat. It checks everything and gets the AI model on its first run — that part takes a few minutes, once.
  4. If Windows shows a security warning, that's normal for any downloaded script, not a sign something's wrong. Most often it's "Windows protected your PC" or "Publisher could not be verified" with a Run button — just click Run (click "More info" first if you don't see Run right away). If instead it says "Smart App Control blocked a file that may be unsafe" — a stricter warning with no Run button at all — right-click Start Lantern.bat → Properties → check the "Unblock" box near the bottom → OK, then try again. Either way, this only affects this one file.
  5. That's it. It opens on its own from here on.

Lantern

A real neural network, trained entirely from scratch — not word-matching, nothing borrowed from anyone else's model. It arrives with a small amount of starter knowledge already trained in; teach it your own below any time to replace that with what you actually want it to know. Training takes real minutes, not seconds — that's genuine, not a bug.

Teach it something

This runs a real training loop — expect a few minutes, not an instant response. Longer or more varied text takes longer.

Its weights are saved on this device after training, so it remembers what you taught it next time you visit — nothing is sent anywhere.

What you're getting

The short version of what makes Lantern different.

  • A real conversation, not a toy. Lantern runs a genuine AI model on your own machine using Ollama — not a scaled-down demo.
  • No account needed for the core chat. No cloud dependency, no data collection for the regular Ollama-based chat — works with no internet connection at all once set up. Lantern 1.0 (an optional model built into Lantern) and the optional voice feature are the two modes that do use the internet — see the Questions page for specifics.
  • Opens like a real app — its own window and icon, not a browser tab.
  • Organized how you like. Multiple conversations, grouped into projects, renamed and deleted freely.
  • Made for one person. The first time it opens, it asks your name and what to call it — and remembers, just for you.

Choosing a model

Lantern comes with llama3.2 already set up, which suits most computers. If replies feel slow (or you want something smarter), here's how to switch — it takes about two minutes.

  1. Open Command Prompt. Press Windows, type cmd, press Enter. A black window opens — that's normal.
  2. Copy the command for your computer from the list below (tap Copy), then paste it into that black window with Ctrl + V and press Enter.
  3. Wait for it to finish. You'll see a progress bar. Most models are 1–5GB, so this takes a few minutes depending on your internet.
  4. Pick it in Lantern. Open Lantern, tap the three-lines icon (top left), and choose your new model from the Model dropdown. It appears within a few seconds — no restart needed.

Very old or very low-end laptop

~2–4GB RAM · struggling with the default

ollama pull llama3.2:1b

The smallest usable option — only worth switching to if the default is genuinely too slow. Answers are noticeably simpler, but it stays responsive where bigger models crawl.

Average laptop

~4–8GB RAM · no graphics card

ollama pull llama3.2

This is the default Lantern already installs, and the best all-round balance for most people — you likely don't need to change anything. If you want stronger reasoning at similar speed, phi4-mini is a good swap.

Newer laptop or gaming PC

8GB+ RAM, or any dedicated graphics card

ollama pull llama3.1:8b

Meaningfully smarter and more capable — the best step up if your machine can handle it. With 16GB+ or a dedicated graphics card you could go bigger still (phi4 is a strong 14B option), but 8B is the reliable sweet spot.

Not sure what you have?

how to check

Press Ctrl + Shift + Esc to open Task Manager, click the Performance tab, then Memory — the number at the top right is your RAM. Start with the default; only switch if it feels slow.

Questions

Worth reading before you download, especially the one about where your data goes.

Why do I need to install Ollama separately?

Ollama is the free program that actually runs the AI model on your computer. Lantern is the interface on top of it. This split is what keeps everything free — there's no company server involved in running the core chat.

Does any of my data leave my computer?

Depends which part of Lantern you're using. The regular chat, powered by Ollama, runs entirely on your own computer — no account, no analytics, no data sent anywhere, and you could disconnect from the internet entirely after setup and keep chatting. Lantern 1.0 is different: it's a separate AI option in the model menu that can search the public web when it doesn't already know an answer, so using it does send your questions to outside search servers. Voice is a third, fully optional mode — off by default — that sends audio to Google's cloud speech-to-text and text to Microsoft's cloud text-to-speech when you turn it on. You'll see a clear notice before voice is turned on, and typing works exactly the same either way.

Can I trust everything it tells me?

Not blindly — no AI model is perfect, including this one. It can sound confident while still being wrong, especially on specific facts, dates, or numbers. Treat it like a knowledgeable friend rather than an authoritative source, and double-check anything that actually matters — particularly medical, legal, or financial questions.

What about email or feedback?

The "Having trouble?" and "Got feedback?" links on the Get Help page just open your own email app with an address already filled in — whatever you write goes directly to whoever maintains this site, the same as any other email you send. It isn't sold or shared beyond that. Skip them if you'd rather not reach out at all. For what happens to your actual conversations, see the question above about where your data goes.

Does this cost anything?

No. Ollama is free, the AI model is free to download, and Lantern is free. The only "cost" is your own computer's processing power while it's generating a reply.

Windows shows a security warning when I try to open it — is Lantern actually unsafe?

No — any script downloaded from the internet triggers some form of Windows warning, regardless of what it actually does. Most commonly you'll see "Windows protected your PC" or "Publisher could not be verified" with a Run button, which is safe to click through. A smaller number of people (with a newer Windows 11 feature called Smart App Control) will instead see "Smart App Control blocked a file that may be unsafe" — this stricter version has no Run button at all. The fix for that one: right-click Start Lantern.bat, choose Properties, check the "Unblock" box near the bottom, click OK, then run it again. Either way, this only affects that one file on your computer — nothing else changes.

Is Mac or Linux as reliable as Windows?

Windows has been tested the most, through real back-and-forth with actual testers. Mac and Linux are newer — the core (the local server, the AI conversation itself) has been directly tested and works the same way everywhere, but the platform-specific setup pieces haven't had the same real-world mileage yet. If something doesn't work, the exact error message helps a lot — nothing is designed to fail silently.

It's really slow to reply — did I do something wrong?

Probably not. The AI runs on your own computer's hardware, so speed depends entirely on your machine — laptops without a dedicated graphics card have to work much harder. Switching to a lighter model (see the Choosing a Model page) usually fixes it. Closing other heavy programs while chatting helps too.

Something's not working right — is that expected?

Possibly! This is early, hand-built software being tested with a small group of friends, not a finished product. If something breaks or feels off, that's genuinely useful to know — see the Get Help page.

Get help

Both of these just open your own email app — nothing is collected by this site.

Having trouble?

Tell me what happened and I'll help sort it out.

Helpful to include: what you were doing when it happened, and the exact wording of any error you saw.

Email me about it

Got feedback?

Ideas, complaints, anything — it comes straight to me.

Send feedback