Nothing leaves your machine.

Lore runs as a local process on your own computer, bound to 127.0.0.1. There is no sign-up, no login and no database anywhere but your disk. Nobody operating Lore can see your wiki, because for this version there is no operator — the copy you run answers only to you.

This is not a promise about what we choose to do with your files. It is that the local build contains no code that could upload them. Everything below describes that build unless it says otherwise.

A hosted plan exists for people who want their wiki on more than one machine, backed up somewhere the machine is not, and reachable by an agent when that machine is asleep. Those are the only things it adds, and none of them are possible without a copy of your wiki living on a server.

So: if you turn sync on, your wiki leaves your machine. It is encrypted in transit and at rest, but it leaves. That is the whole trade, stated plainly, and it is off until you deliberately switch it on. Nothing about the free build changes when hosted plans exist — it has no sync code in it at all.

No tracking scripts, no error reporting service, no usage pings, no feature flags phoning home. The application has no analytics dependency of any kind, and the only external addresses written anywhere in its source are the two links in the footer.

Lore does record how your agents use your wiki — which pages get read, which searches come back empty. That log is a file in ~/.lore on your machine. It is written locally, read locally, and never transmitted.

Everything Lore keeps for itself lives in ~/.lore, deliberately outside your wiki, so a git diff of your notes stays clean: the vault path you chose, the write journal, the verification ledger, the usage log, and any model you downloaded.

Inside your wiki it writes two things, both only when you ask: pages you create or edit in the app, and an AGENTS.md index when you press the button — fenced with markers so anything you wrote in that file is preserved.

Semantic search, if you turn it on, downloads a small embedding model from Hugging Face the first time and caches it in ~/.lore/models. Your pages are not sent anywhere — the model comes to your machine and the embedding happens there. After that download it works offline.

A local language model, if you have Ollama running, is reached at 127.0.0.1:11434. That is your own machine. Lore does not bundle a model, does not install one, and has no cloud model provider configured.

Lore can be opened from your phone on your own network. It is off unless you switch it on, and when you do it issues a random token stored with owner-only permissions. The connection is plain HTTP over your local network — there is no TLS, so a hostile network could read the token. Treat it as a convenience on networks you trust, not a security boundary.

The page you are reading is a static marketing site with no analytics on it. It is served by a hosting provider that keeps its own request logs, in the ordinary way that every web server does — IP address, page, timestamp. Nothing on this site sets a tracking cookie, and none of it is connected to the application.

Do not take it on trust. The source is public: search it for a network call and see what you find. Or run Lore with the network off — everything except the one-time model download works exactly the same, which is the shortest proof available.