Skip to content

FAQ

Graph content (concepts, definitions, examples, notes, relation structure) and review progress (FSRS stability, ratings, due dates) are stored locally, in IndexedDB on the web app or as plain JSON files on disk for the desktop app, and never leave your device. Mentor chat history lives only in memory for the current panel session, and AI endpoint API keys are stored locally and sent only to the endpoint you configured.

The only things that can leave your device:

  • AI mentor prompts and chat history, if you enable the mentor with a remote endpoint. A local Ollama endpoint sends nothing off your machine.
  • Opt-in telemetry, off by default.
  • The desktop app’s version check against GitHub Releases, a plain request that carries no graph or usage data.

What changes between the web app and the desktop app?

Section titled “What changes between the web app and the desktop app?”
AreaWebDesktop
Graph storageIndexedDB onlyPlain JSON files on disk (authoritative); IndexedDB mirrors the active project
ProjectsSingle implicit workspace, no switchingMultiple named project folders: create, open, switch, remove
File import/exportFile System Access API or browser downloadNative save/open dialogs
Native app menuNoneA real File/Edit/View/Window/Help menu bar
Auto-updateNot applicableChecks GitHub Releases on launch, can install and relaunch
External file changesNot applicableA file watcher detects edits made outside the app (see Troubleshooting)

Everything else, the canvas, the Inspector, Review, and the AI mentor, works the same on both.

It doesn’t, if you run a model locally. The default setup points at Ollama on http://localhost:11434/v1, which needs no key. An API key is only required when you point the mentor at a hosted provider that authenticates requests (most OpenAI-compatible APIs); Nesso sends it as a bearer token to that endpoint only, never anywhere else. See AI mentor.

Where do I report a bug or request a feature?

Section titled “Where do I report a bug or request a feature?”

Open an issue on GitHub. The code is MIT-licensed and contributions are welcome.