What Offline Base is building, and what you can use today

We're building three products that run AI on hardware you own: Base Drive, Base Router and Decision Tree. None can be ordered yet, and several free demos work today.

Offline Base LLC is a small company built on one idea, written at the top of our About page (opens in a new tab): "AI has become part of daily life, so it should live on hardware you own, under rules you set."

Our home page puts it in six words: Making Local AI Easy for All (opens in a new tab). Local AI means the model runs on your own computer, so what you type stays on the machine and the model keeps answering with the Wi-Fi off. Today that takes more setup than most people will put up with, and our products are meant to remove it.

Three products and where each one stands

The shop page (opens in a new tab) names Base Drive "the first product to launch," and checkout will run through Shopify once ordering opens.

Base Drive's installer copies the AI onto your laptop, so the model keeps running with the drive unplugged. The plan puts two models on every drive: Liquid AI's LFM2.5 1.2B and Google's Gemma 4 E2B. Our Base Drive post walks through the rest.

Base Router puts the same private AI on dedicated hardware that stays on for the whole household. Its price and speed figures are targets. Our Base Router preview covers the design.

Decision Tree answers yes-or-no, pick-one, pick-any and score questions, and gives a probability with every answer. More in our Decision Tree post.

Base Connect, a planned connector between your local memory and ChatGPT or Claude, has its own integrations page (opens in a new tab), and nothing on it ships yet. We are still designing optional outside help, such as web search you approve first.

Four promises on every product

We hold every product to four promises from our About page:

• Private. "Nothing leaves the box without your explicit approval, and you can turn outside access off entirely."
• Local. It "works offline, and does not require another account or a usage meter."
• Fine-tuned. We tune it to your documents, routines and use case.
• Packaged. "It just works." You plug it in, open it from your devices and start using it, with no technical setup.

Honest before impressive

That heading comes from the "How we work" section of our About page, which opens: "We'd rather under-promise." Every product page tells you, before you buy, that a small local model trails the largest cloud systems.

Our products send no telemetry, push no quiet updates and make no hidden cloud calls. If a task ever needs outside help, you approve it first and a record is kept.

Our terms (opens in a new tab), last reviewed September 6, 2026, call every specification, price, model list, speed figure and date we publish "a current target." Read the prices in this post with that in mind.

What you can use today

These are free and need no purchase, and all were live on October 14, 2026. The demos are rate limited and can go offline, and /try stops after three messages.

• /try (opens in a new tab): chat with Liquid LFM2.5 1.2B running live on one of our Jetson Orin Nano Super computers, the small NVIDIA board Base Router is built on. The chats aren't stored.
• /open (opens in a new tab): chat with 10 open-weight models (GLM 5.3 Flash by default) with no account and nothing saved. Model calls go only to providers with zero data retention. Web searches the model runs and PDF parsing you opt into fall outside that guarantee, as our zero data retention post explains.
• /fleet (opens in a new tab): live numbers from our own rack of Jetson Orin Nano boards. At 11:01 UTC on October 15, 2026, 4 of 5 boards were online, and the fleet had served about 4.34 million tokens across 6,691 requests since September 30, with none sent to a cloud model.
• Gemma Gemini (opens in a new tab): our spinoff, a free Gemini-style chat with no account. Each message goes to Gemma 4 E2B on our Jetsons or to Gemini 4 Argon through OpenRouter, and every reply shows who answered and why.
• Sell your old computer (opens in a new tab): send a photo of a desktop, laptop, workstation or graphics card from about 2014 on. We are still setting our buy prices, so any number the page shows is an estimate. A person on our team quotes the real figure before we collect anything.
• The Offline Base Letter (opens in a new tab): our newsletter on Substack, with one issue so far, "Intro to Offline Base" from August 30, 2026.
• @offlinebase on X, Instagram and YouTube: the links sit in the footer of offlinebase.com (opens in a new tab).

Three things you can try on offlinebase.com without an account walks through the demos, and How our computer buyback works covers the buyback.

The five beliefs we started on

We first published these in "Intro to Offline Base," and the About page keeps them.

1. "Local, open, small models are good enough." A Stanford preprint, Intelligence per Watt, found local models answered 88.7% of one million real single-turn queries. More in What a million real questions say about small local models.
2. "Running models locally answers the data-center problem." Everyday questions answered by a small model on hardware you already own skip the data center and the electricity it uses. More in The energy case for local AI, with our caveats.
3. "Teaching everyone data privacy is a lost cause (sort of)." People will keep pasting sensitive things into the nearest chatbot, so we want local to be the default, with heavier work sent to bigger models only when the owner approves. More in Why we make local the default for private AI.
4. "No one gets left behind in an AI-native world." We are young, a lot of our friends want nothing to do with AI, and Pew found in 2026 that 55% of US adults under 30 are more concerned than excited about it. More in Building AI for people who are wary of AI.
5. "Open-source models are more than good enough." Frontier models are tools to call on when a question needs them, and in our words, "Routing is the name of the game." More in How a decision model picks between our Jetsons and the cloud.

Try a model now, or read up on Base Drive

The quickest way to watch a small model answer on hardware like ours is /try (opens in a new tab). For Base Drive, start with the Base Drive page (opens in a new tab), which explains how setup checks SHA-256 checksums before it copies anything.