Ollama support is Rixoffice's strongest privacy statement: choose a local model and the entire AI stack — documents, prompts, agent reasoning and responses — never leaves your machine. It is also the only provider that works with no internet at all, which makes it the answer for flights, secure environments and anyone who simply prefers their files to stay home. Setup takes minutes, and after that the composer behaves like any other provider.
Why run models locally
Three reasons come up again and again:
- Privacy. Nothing is transmitted. Your document is processed on your device, the prompt never crosses a network, and the answer is generated by hardware you own. There is no third party in the loop at any point.
- Offline capability. Every AI feature keeps working with the network cable pulled. Combined with Rixoffice's offline-first editing, a laptop with Ollama is a complete, connected-nothing office.
- No metering. Local models consume electricity, not credits. There is no per-action cost, no quota to watch and no monthly ceiling to plan around.
The trade-off is hardware: local models are smaller than the frontier cloud models, so quality on complex tasks depends on what your machine can run. For a great deal of everyday office work, the gap is smaller than people expect. Rewriting a paragraph, summarizing a section, explaining a formula and drafting a routine email are all well within reach of a mid-size model on a modern laptop, and those tasks make up most of what an office AI actually does in a day.
There is also a cost angle beyond metering: a one-time model download replaces a recurring per-action charge. If your AI volume is high and steady, the electricity bill for a local model can be a rounding error compared to any metered alternative.
Setting up Ollama with Rixoffice
- Install Ollama from ollama.com — it runs on Windows, macOS and Linux.
- Pull a model from a terminal. For a general-purpose start: ollama pull llama3
- Leave Ollama running in the background. It listens on localhost by default, which is exactly where Rixoffice looks.
- Open Rixoffice Settings, go to the AI provider section and choose Ollama.
- Select the model you pulled from the model list and test the connection.
- Open any document and try the composer. If you want proof it is local, disconnect your network and run the same request.
Setup is a one-time job. After that, Rixoffice remembers the provider and model, and the composer behaves exactly as it does with any cloud provider — same agents, same skills, same block-level edits.
Choosing a model
The right model depends on your hardware and your work:
- General writing and editing: a mid-size general model such as llama3 is a solid default on machines with 16 GB of memory.
- Long documents: prefer the largest model your machine comfortably runs; context handling improves with model size, and long contracts or reports are where small models struggle first.
- Fast chats and simple summaries: a smaller model responds quicker and is often all a task needs.
You can pull several models and switch between them in Settings. Because each has its own strengths, many people keep one large model for heavy generation and one small model for quick questions. Switching costs nothing and takes seconds, so experimenting is cheap.
What stays on your machine
With Ollama selected, the full chain is local:
- Your document is read by the agents on your device.
- Prompts and agent reasoning happen on your device.
- Model weights live on your disk, downloaded once via Ollama.
- Answers are generated on your device.
The only network activity left is what you choose: updates, sign-in, cloud projects and web search still use the internet when you want them, but none of them are required for AI. Pull the cable and the editors and the AI keep working; plug it back in and sync resumes. The full privacy picture across all providers is in AI privacy.
Worth knowing as well: model weights are downloaded once through Ollama and cached on disk, so even setup does not create a standing dependency. After the pull, the workflow has no cloud touchpoint unless you invoke one deliberately.
Combining local and cloud
Ollama is a provider, not a mode you are locked into. On the Pro API & Local plan you can keep Ollama for sensitive work and switch to built-in Rix AI or a BYO key when a task needs a frontier model. The agents, skills and block-level editing are identical across all of them — only the engine changes.
A common pattern: draft and iterate on local models where the volume is high, then run one final review or translation on a cloud model where quality peaks. You choose per task, not per subscription.
Where local models shine
Some workflows are a natural fit:
- Drafting and editing confidential contracts, medical or legal notes where nothing may leave the premises.
- Working on planes, trains or anywhere without reliable connectivity.
- Heavy experimentation and bulk edits where per-action metering would add up.
- Field work — audits, site visits, ships, camps — where the nearest data center is a very long way away.
What to explore next
- AI privacy — the full privacy model across all providers.
- Offline mode — everything Rixoffice does without internet.
- AI provider setup — where Ollama sits among the cloud options.