Octopi / Guides / Sovereign AI
Explainer
Sovereign AI, on your own Mac
Sovereign AI means you control the AI you rely on: the model runs on hardware you own, its weights are yours to keep, and your work isn’t sent to an AI company to be processed. Octopi works that way by default. Here is exactly what runs where, including the parts that do go online.
What makes AI sovereign
Your hardware
The model runs on your Mac, through Ollama, the open-source engine that serves it — or on another computer you run, if you point Octopi at it.
Your model
An open-weight model — Qwen by default; Llama, Mistral and Gemma are offered too — downloaded once and kept on your disk. You can switch or remove it at any time.
Your data
Your chats, the documents you attach and your spreadsheets are processed on your Mac. The chat conversation itself is not saved.
| In Octopi | Where it happens |
|---|---|
| The AI model | On your Mac (Ollama), unless you choose another Ollama computer you run |
| Your chat conversation | On your Mac — and it isn’t saved |
| Documents you attach | Read on your Mac; their passages go only to your own Ollama model |
| Your spreadsheets | On your Mac — never uploaded |
| Web and CRM look-ups you ask for | Online: Octopi visits those sites and your signed-in CRMs for you |
| Claude (optional) | Only if you switch it on in Settings, with its own consent screen |
| Usage & diagnostics | Sent to Octopi by default, linked to your download email — switch off in Settings |
| Updates | The app checks GitHub for new versions |
Sovereign doesn’t mean offline: Octopi goes online to do what you ask, and says where.
How Octopi fits in
The model is on your Mac
Out of the box, every answer is written by a model running on your own computer. No AI subscription is needed.
A model you can keep
The default model, Qwen2.5-7B-Instruct, is published under the Apache 2.0 licence. Other models’ licences vary — check each one’s terms.
Your own Ollama comes first
If Ollama already runs on your Mac, Octopi uses it. If not, it can install a pinned, checksum-verified Ollama release — and stops it when you quit.
Claude only if you ask
Planning or reading with Claude is off until you turn it on, with your own Anthropic key or Octopi credits. Each option has its own consent screen.
Questions
Is Octopi fully offline?
No. The AI model runs on your Mac, but Octopi goes online to do what you ask — searching the web, reading pages, using your signed-in CRMs — and to download the model once, check for updates and send usage and diagnostics (which you can switch off).
Does an AI company see my data?
Not by default: the model is on your Mac. Only if you turn on the optional Claude features in Settings is anything sent to Anthropic, and each feature’s consent screen says exactly what.
Which models can I use?
Qwen models by default (Light, Balanced, Newest and Advanced sizes), with Llama 3.1, Mistral and Gemma 2 also offered, and any model you have already installed in Ollama. Octopi’s prompts are tuned for Qwen, which is why it is the default.
Can I run the model on my own server?
Yes. Point Octopi at an Ollama server you run on another computer, and the chat says your messages go there. The chat refuses a model that Ollama runs in its own cloud.
What does Octopi send to its own servers?
By default, run counts, outcome categories, setup progress, your Chrome version, crash counts and scrubbed error details, linked to the email you downloaded with. Your spreadsheets, contact lists and prompts are not sent, though a scrubbed error line can still name a page or field. Turning it off in Settings unlinks and stops it.
Do I need an account or a subscription?
No. Octopi is free and needs no account. Octopi Cloud credits are an optional extra for the Claude features.
Related
Keep your AI on your own Mac
Download Octopi and run an open-weight model locally — free, with the cloud strictly optional.
Download Octopi free