A Lightweight Coding and Science Helper, Without Any Local Setup

So, what exactly is RNJ-1?

RNJ-1 is a compact AI model built by Essential AI specifically for coding, math, science, and basic automated tool use. It was trained from the ground up rather than adapted from a larger model, and it's designed to handle a good chunk of conversation or code at once. Think of it as a focused specialist rather than a broad generalist -- it's aimed at programming help, solving science and math problems, filling in missing pieces of code, and small experiments where the AI completes a task on its own with some outside tools.

Through Terminal Glass's cloud connection, you get access to the standard version of RNJ-1. A smaller, downloadable version also exists if you'd rather run everything on your own equipment through a fully private setup. Whichever way you go, it's worth setting clear rules upfront about what code and documents are okay to send out over the internet.

"Do I need to buy a fancy graphics card?"

This is really the question most people are trying to figure out, even if they don't phrase it that way.

If you try to run these AI models without a proper graphics card (the same kind used for video games, but scaled up), your computer will fall back to using its regular processor. That technically works for the occasional light question, but it becomes painfully slow for regular use. A lot of people learn this the hard way after renting a cloud computer that turns out to be either too slow or surprisingly expensive.

If you'd rather not deal with picking out graphics cards, comparing specs, or troubleshooting driver issues at all, Terminal Glass Cloud is the easier path -- you get a capable AI through a simple, familiar chat window, and someone else worries about the hardware.

On the other hand, if you want everything running on equipment you personally control, you can rent a graphics-card-equipped server (through AWS, Digital Ocean, or your own machines) and install NoCloudGPT on it yourself. In that setup, you're the one responsible for picking the right graphics card and keeping it running smoothly.

And if your organization needs everything to stay completely inside your own walls from the very first day, then a NoCloudGPT Air Gapped setup on your own hardware is the way to go.

There's no single "correct" choice here. It really comes down to your budget, how sensitive your data is, how fast you need answers, and whether you want to own your equipment long-term.

Is this actually for you?

How this actually works, step by step

You'll interact with everything through OpenWebUI -- a clean, simple chat window running on your own computer (or you can use the command line, if you prefer). Your chat history, user accounts, and any documents you've indexed for search stay on hardware you control. But the moment you send a question, it -- along with any code snippets or document excerpts your search tool pulls in -- travels out to Ollama's cloud servers to be answered, and the reply streams back to your screen.

To use the cloud version, you'll need to sign in with an Ollama account on your computer first. RNJ-1's cloud usage is priced in a moderate tier -- generally reasonable for a model this size, though repeated back-and-forth with outside tools can still add up over time.

ollama signin
ollama run rnj-1:8b-cloud

What leaves your computer: your typed questions, code snippets, any document text your search tool pulls up, and any back-and-forth with outside tools. This model only reads and writes text -- it doesn't handle images. Speed: includes some delay for the round trip to the cloud and back. Cost: based on how much you use it. When you're ready, the same model can run fully on your own hardware through a private setup instead.

When you should probably choose something else

• You have regulated data or strict rules about where data can be stored -- Choose a fully private setup running RNJ-1 (or another coding model) entirely on hardware you control. This cloud version sends your questions and code out to Ollama's servers.
• You use it every day, and you already own decent hardware -- The downloadable version of this model is small enough to run on many modest servers, and may cost less over time than paying for cloud usage continuously.
• You need to rework an entire codebase or run heavy automated coding tasks -- A bigger, more capable coding-focused model such as Qwen3-Coder Cloud or Devstral Small 2 Cloud may be a better fit than this smaller model.
• You want a general-purpose assistant for everyday writing or conversation -- RNJ-1 is built for coding and technical problem-solving. Give it clear instructions if you want to use it for non-technical writing or casual conversation.
• You need the exact same, unchanging version of the model for audits or records -- A private setup with a fixed version gives you more control and predictability than a cloud service that can update behind the scenes.

What we recommend

A smaller, downloadable version of this model is also available for anyone who wants to run it entirely on their own equipment through a fully private setup. What's available can change over time -- double-check the current options at ollama.com/library/rnj-1 before committing.

The legal fine print, in plain terms

RNJ-1 is published by Essential AI under the Apache 2.0 license, which allows commercial use as long as you follow its standard terms. Using the cloud version requires an Ollama Cloud account, and you're billed based on how much you use it -- usage limits and pricing tiers apply and can change. Using Ollama's cloud service also means agreeing to Ollama's own separate terms -- worth reviewing both before relying on this for business purposes.

Official listing: ollama.com/library/rnj-1 · More information: Essential AI's RNJ-1 research post