Settings: Privacy
"What leaves this computer when you talk to the assistant, and what is kept." The switches decide what is sent and stored; the boxes below say it plainly; the last part manages keys. Why the app is built this way: Privacy.
Switches
| Switch | Default | What it does |
|---|---|---|
| Send field values to the model | on | "Prompts, seeds, model names in the nodes. Values that look like keys are always hidden." Off: the model sees the graph without the values. |
| Keep chat history in the project | on | "Off: new messages stay in this window only and are not written." |
| Keep a log of the conversation with the model | on | Every question is recorded with what was sent and what came back. It stays on this computer, holds no keys or tokens, and old records go by themselves. See The conversation log. |
| Adult content (18+) | off | See Adult content below. |
| Look up models on the internet during sync | off | For files Civitai's description does not cover: the fingerprint goes to civitai.com, then the name to huggingface.co. At most one request a second and 200 files per sync. While off, only Find on the internet in a card does it. |
| Check for new versions | off | Once a day the app asks the project's releases page on github.com whether a newer version exists. The question carries only the app's name and version. The app downloads nothing by itself. |
Adult content
The switch Adult content (18+) is off in a new profile. "On: the previews and example descriptions an author marked as adult content are shown, and the lessons for adults are open. Off: they stay closed, and a model card still shows them for one file when you ask."
Turning it on asks first: "Turn on adult content?", with the rules, Read the content rules, I am 18 or older and Leave it off. See Content rules; Help ▸ Content rules opens them too.
What is sent to the model
A box lists it: your message, the graph (nodes, inputs and outputs, links, positions), what is selected and on screen, your recent actions on the graph, a summary of your catalog; when the assistant looks at a model, the description saved next to it and what Civitai or Hugging Face said about it. Then "Field values: sent." or "Field values: not sent." and "Never sent: model keys, the server access token, the model files themselves and the folders they lie in." See What is sent to the model.
A second box says where else the app connects, and when: the assistant's web search (DuckDuckGo from this computer, or the cloud provider's own search), the sign-in page of your MCP server (the app asks first when it is on another site), a check of a few known ports on this computer, and Find on the internet in a model card (civitai.com, huggingface.co and their picture servers).
Where keys are kept
A table, Key and Where it is now, with a row per key: "OpenAI key", "Server access token". The place is "In memory until the app closes", "Encrypted in the system keychain", "In the app folder, unencrypted (your choice)" or "Not held".
- Forget all keys and tokens removes every model key, every server access token and every browser sign-in, from memory, the system keychain and the app folder. The app asks first and lists what goes.
- System keychain, Which keychain the app uses: Detect automatically, GNOME “Passwords and Keys”, KWallet 6, 5 or 4. "Now: GNOME “Passwords and Keys”. Keys are saved encrypted." A change applies after the app restarts.
The picture cache
The app keeps preview pictures and clips of your models so that cards open at once. The line says how many files and how much room they take; Clear the picture cache removes them. They are also cleared when the server address changes.
Chat history
Clear chat history removes the stored messages of this project after a question ("Messages of this project to remove: 24. Explanations saved on nodes stay."). "There is no chat history to clear." when there is none.