Three things you can try on offlinebase.com without an account

Three parts of offlinebase.com work today with no account and no payment: the /try chat on a Jetson, the /open chat with 10 open-weight models and the "Ask a small model to…" rewriter. As of October 2026, every product we are building is still coming soon.

For the products themselves, see What Offline Base is building, and what you can use today.

1. /try: a small model on a Jetson

offlinebase.com/try (opens in a new tab) is a live chat with one model, Liquid AI's LFM2.5 1.2B (opens in a new tab). The page says it runs on an NVIDIA Jetson Orin Nano Super that we operate, "the small NVIDIA computer Base Router is built on." Base Router, the always-on AI box we are designing explains the product, and Inside the Jetson Orin Nano Super, the computer behind Base Router covers the board.

The demo shows what AI on a small box you could own feels like: replies stream at the pace the board really manages, about 38 tokens a second. That is a representative figure for this model on this board, so any single reply can land a little above or below it.

The page says: "These chats aren't stored." It also lists its limits:

• One model. /try offers only LFM2.5 1.2B, so everything on the page runs on the box.
• Three messages. A conversation ends after three of your messages.
• It can go offline. The page calls itself "a demo on one small computer: rate limited, and it can go offline." When that computer is down, the chat is down.

LFM2.5 1.2B handles text only, and it is one of the two models our site lists for Base Drive. Ask it for a short rewrite or a quick draft and watch the words arrive. Models can be confidently wrong, so check anything that matters.

2. /open: 10 open-weight models, nothing saved

offlinebase.com/open (opens in a new tab) is a free chat with 10 open-weight models. Its header reads "Open models. No account. Nothing saved." The models' weights are public, so anyone with enough hardware can run them, though the biggest ones need a server rack. On /open they run in the cloud, through OpenRouter.

The lineup as of October 2026:

Sizes are total parameters. Several of these are mixture-of-experts models that use only a fraction of themselves per word: GLM 5.3 Flash, for example, uses 18B of its 320B.

What the chat can do:

• Reasoning panel. Models that think before answering stream that thinking into a panel that collapses once the answer starts.
• Tool calling on every model except Nemotron 3.5 Lightning, which therefore can't search the web or make files either. One tool looks up Offline Base product facts from the same data as the rest of the site (the Base Router speeds it returns are speed targets from the preview).
• Image reading on the six models marked above.
• Web search, decided by the model. When a model decides a question needs the web, it calls OpenRouter's web-search tool and lists its sources.
• PDF reading. When you attach a PDF, OpenRouter's file-parser plugin turns it into text.
• Text files and Markdown export. A model can write a text file (notes, code, a CSV), and you can download the conversation as Markdown. Your browser assembles both from the reply, so we never save them on our servers. A file stays in your browser until you download it.

/open is rate limited per IP address and capped per reply.

Where the privacy line sits

Every request from /open carries OpenRouter's zdr flag. OpenRouter's documentation (opens in a new tab) says a request with that flag "will only be routed to endpoints that have a Zero Data Retention policy." If a model loses its zero-data-retention endpoint, the request fails.

Two features sit outside that guarantee: a web search sends its query to a search provider, and an attached PDF goes through the file parser. The same OpenRouter page (opens in a new tab) draws that line: ZDR enforcement "only applies to provider routing for inference requests," which leaves out plugins and tools such as web search. The PDF leg only happens when you attach a PDF, so that part is your choice.

Zero retention is still a promise made by a third party. What zero data retention means, and where it stops covers the details.

3. "Ask a small model to…" on most pages

At the bottom of most pages on offlinebase.com sits a button labeled "Ask a small model to…". Open it and ask for the page rewritten. The quick choices are translate (Spanish, Japanese, French, Chinese or German), "make it funny" and "ELI5", or you can type your own request of up to 200 characters.

A small model on our own Jetsons does the rewriting. By default that is Liquid AI's LFM2.5-VL-3B (opens in a new tab), one of the models you can see loaded on our live fleet page (opens in a new tab).

Here is what happens when you ask:

1. The page sends only the visible text, in small batches.
2. The model rewrites each piece, and the answers go back into the same spots. Buttons, links and layout stay put.
3. Rewritten words appear in Comic Sans (or the lookalike Comic Neue where Comic Sans isn't installed) so you can tell them apart from ours, and a bar along the bottom offers "Show the real text."
4. Your request follows you from page to page.

One thing stays fixed: the name "Offline Base." The prompt pins it and each reply is checked for it. If a piece comes back missing or mangled, the page keeps our original wording for that spot.

We leave the rewriter off /privacy, the Base phone app's policy, because that wording has to stay exactly as written. The /try and /open chat pages don't have it either. The rewriter is rate limited per hour, and a 3B model makes mistakes, so read a translated page as a rough guide.

The terms that cover all three

Our terms (opens in a new tab) say the demos "are rate limited, they can be slow, they can be switched off without notice, and we make no availability commitment for any of them." They also say: "Anything you paste into the demos on this site leaves your machine by definition."

That second line applies to all three. /try sends your words to our demo computer, /open to a cloud provider and the rewriter to our Jetsons. Keeping your words on your own computer is the job of the products we are building. Base Drive is planned at $19.99, paid once, and cannot be ordered yet. Base Drive, private AI for the laptop you already own explains how it works.

Two sibling sites

• tryingopen.com (opens in a new tab) is our other free chat with open-weight models.
• Gemma Gemini (opens in a new tab) is a free Gemini-style chat with no account. Each message goes either to Gemma 4 E2B on our Jetson Orin Nano boards or to Gemini 4 Argon through OpenRouter, and every reply shows a route trace saying who answered and why. Gemma Gemini, a free chat that shows which model answered and why tells that story.