Privacy

This page lists every place NodeWise Studio can contact, when it does so, on whose action, and what goes there. Nothing else leaves your computer.

#What the app never does

• No telemetry. It does not count, measure or report how you use it.
• No crash reporting. Errors go to a log file on your computer only.
• No update check unless you ask for it. Settings ▸ Privacy ▸ Check for new versions is off by default (see below).
• No account. It needs no sign-up with the project.
• No spell checker downloads. The spell checker of the window is switched off, so Chromium does not fetch dictionaries.
• No fonts or scripts from the internet. Everything the window shows is inside the app.

#Every place the app can contact

#Your ComfyUI

• When: when you check or read ComfyUI (Check ComfyUI, Read now, Sync ▸ Sync now), on the sync timer if you switched it on in Settings ▸ Sync, and when a model card shows a picture ComfyUI has for a model.
• Where: only the ComfyUI address you saved in Settings ▸ Connection. Find on this computer asks only this computer (127.0.0.1, ports 8188 and 8000).
• What goes there: read-only GET requests, through one gate that refuses every other route: the list of nodes (/object_info), the versions (/system_stats), the model folders and files, the metadata inside a .safetensors header (/view_metadata), a model's preview picture, and, when ComfyUI-Manager is installed, its list of installed packs. Never a prompt, a queue, an upload, an install or a setting. No password is sent: ComfyUI has none.

#Your MCP server (optional)

• When: as for ComfyUI, if you set an MCP server address in Settings ▸ Connection; and when you choose Find on the internet in a model card (the server computes the file's fingerprint).
• Where: only the MCP server address you gave, and a ComfyUI address it names if that is on this computer too, or is the address you saved.
• What goes there: calls of the few read tools the app knows, with fixed arguments: the node, pack and model lists, the files next to a model (description, example picture), a pack's pyproject.toml and README, a model file's fingerprint. The access token or the sign-in, if the server needs one. A server whose tools the app does not know is not called at all.
• Sign-in through the browser: your system browser opens the server's sign-in page; the app waits for the answer on a local address of this computer (127.0.0.1).

#The model that answers (the assistant)

• When: only when you send a message to the assistant, use Ask about…, or choose Check connection in Settings ▸ Assistant.
• Where: the one model you chose:
  • a cloud provider: OpenAI (api.openai.com), Anthropic (api.anthropic.com), Google (generativelanguage.googleapis.com), or the address you set for that provider;
  • your own server (Ollama, LM Studio, vLLM or another OpenAI-compatible server) at the address you gave.
• What goes there: your message; the graph (nodes, their 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. Field values (descriptions, seeds, model names in the nodes) go only while Settings ▸ Privacy ▸ Send field values to the model is on (it is on by default); values that look like keys are always hidden.
• Never: model keys (except the provider's own key in its own request header), the server access token, the model files and the folders they lie in.
• A key goes only to the address it was pasted for. If you change the address, the key is not sent there until you paste it again.
• A server on your computer: while a request runs, the app also asks an Ollama server which context window it loaded the model with (/api/ps), so a request that does not fit can be noticed.

#Web search and web pages (the assistant)

• When: while the assistant answers, if your catalog cannot answer and the assistant decides to search or to read a page. Settings ▸ Assistant ▸ Allow web search and Allow reading web pages: both are on by default; switch them off to keep the assistant away from the internet.
• Where: with a cloud model that has its own search (OpenAI, Anthropic, Google), the provider searches on its side. Otherwise the app asks DuckDuckGo (html.duckduckgo.com, lite.duckduckgo.com) from this computer. Pages are opened from this computer, only public http/https addresses on the usual ports, and only addresses named in the conversation. Your ComfyUI, your MCP server and addresses inside your network are never opened this way.
• What goes there: the search words the assistant chose; a normal request for the page. DuckDuckGo and the site see your IP address, as with any browser. With the provider's own search, the provider sees the search words; the sources it cites are not added to the pages the assistant may open, so an address that carries your words in it is never fetched by the app.

#Model descriptions from Civitai and Hugging Face

• When: when you choose Find on the internet in a model card. During sync only if you switched on Settings ▸ Privacy ▸ Look up models on the internet during sync (off by default).
• Where: civitai.com and huggingface.co, and their picture hosts (image.civitai.com and the Hugging Face file hosts) for an example picture.
• What goes there: for Civitai, the fingerprint (SHA-256) of the model file; for Hugging Face, the name of the file. Nothing else. A picture marked as adult content is not downloaded unless you choose to see it.

#New versions (only if you switch it on)

• When: only while Settings ▸ Privacy ▸ Check for new versions is on (off by default); at most once a day, and an hour later after a failed try.
• Where: GitHub (api.github.com), the project's latest release.
• What goes there: a request that names the app and its version; GitHub sees your computer's address, as any site does. The answer is kept in update-check.json in the data folder. The app downloads nothing: a line in the status bar offers the release page and the download, and you choose.

#Finding servers on this computer

• When: when you choose Find local servers in Settings ▸ Assistant.
• Where: this computer only (127.0.0.1 on the usual ports of Ollama, LM Studio, vLLM and the llama.cpp server: 11434, 1234, 8000, 8080), and your MCP server, if it lists model servers. Find on this computer in Settings ▸ Connection asks ports 8188 and 8000 the same way. Nothing leaves this computer.

#A proxy (only if you need one)

• Requests to cloud models and the web search follow your system's proxy settings. Settings ▸ Assistant ▸ Network ▸ Proxy (optional) takes a proxy the system does not know: http://, https://, socks4:// or socks5://, a host and a port, never a user name or password.
• Your own server on this computer or your home network, your ComfyUI and your MCP server on this computer are always reached directly, not through the proxy.

#Your browser

The app opens your system browser, never a window of its own, for: signing in to an MCP server, Help ▸ Project website, Help ▸ New versions, Help ▸ Report a problem…, and Open ComfyUI on the card "Run it in ComfyUI" (your own ComfyUI address). Report a problem… first shows you what the new issue will contain (the versions of the app and the system, nothing else).

Signing in to an MCP server opens the sign-in page that server names. When that page is on another site than the server, the app asks first and shows both names.

Two download pages are opened there only when you click their buttons in the No ComfyUI yet window: Open the ComfyUI download page (https://www.comfy.org/download) and Open the Ollama download page (https://ollama.com/download). The app itself sends nothing to either site.

#What stays on your computer

• The database with settings, the catalog, the history of the graph and the chats (Settings ▸ Privacy ▸ Keep chat history in the project can switch the chats off).
• The log and the conversation log (Settings ▸ Privacy ▸ Keep a log of the conversation with the model). Neither holds keys or tokens.
• Saved keys: in the system keychain, or, only if you choose it where there is no keychain, in a file in the app folder readable by your user only.
• Your own lessons: one file per lesson in the lessons folder.
• recent-files.json: the last 10 files you opened or saved, and the file each graph came from. Paths only, never the contents.
• graph-start.json: a mark for the safe start ("drawing", "drawn"), so a graph that hung the window is not drawn again without asking.
• The picture cache of model cards (model-previews). Settings ▸ Privacy shows its size and has Clear the picture cache; it is also emptied when the server address changes.

Copy diagnostics and a saved conversation log keep no address of your own servers: only the scheme, where the server is ([this computer], [local network], [internet]) and the port. Public services (the providers' APIs, Civitai, Hugging Face, DuckDuckGo) stay named.

Where each file lies: The data folder. How to remove everything: Removing.

#See also

• Trust
• Security
• ComfyUI is only read
• What is sent to the model
• Settings: Privacy