Privacy
Local AI runs model inference through Ollama on your computer by default. Local tools return results to the client that called them, including model answers and selected source excerpts. A cloud agent can include those results in requests to its cloud provider. Local inference does not guarantee that returned content stays on this computer; the client’s configuration and data policy govern that transfer.
This page lists the connections Local AI itself makes to another computer, what starts them, and what they send. Connections made by a cloud client are controlled by that client.
Local processing and storage
- Model requests. The broker is the only part that talks to a model, and by default it
talks to Ollama at
http://127.0.0.1:11434/, on this computer. A broker you start by hand with another address sends model requests to that address. Results returned to a client can then be forwarded by that client, as described above. - The code index. It is built and kept under
%LOCALAPPDATA%\LocalAi\repositories. - Usage reports and model telemetry. They are written under
%LOCALAPPDATA%\LocalAi\telemetryand read only bylocalai telemetry. Model telemetry refuses any field named for a prompt, an answer, content, an image, a path or a secret. Nothing uploads either of them.
Connections to other computers
| What | Where to | When | What is sent |
|---|---|---|---|
| Finding and downloading a release | github.com, then api.github.com if the first answer does not name a release |
when the installer opens, unless you pointed it at a folder; again before it installs; on localai update |
anonymous HTTPS requests for the release’s files |
| Model sizes | registry.ollama.ai |
when the installer checks the computer | the name and tag of each model it can offer |
| Update check | github.com |
off by default; when you turn it on, about once a day | anonymous manifest/signature requests by default; the separately selected GitHub CLI authentication mode uses the signed-in account |
| Prerequisites | the sources of winget and npm |
only for the prerequisites you tick in the installer | whatever winget and npm send |
| Models | the model registry Ollama uses | when you choose models in the installer, run localai model pull, or approve the local_models_sync tool |
whatever Ollama sends |
| GitHub CLI access | github.com, signed in as you |
installer release access requires explicit selection/consent; update checks require their own explicit GitHub CLI authentication setting | whatever gh sends |
Anonymous update checks send no account authentication or Local AI usage identifier. Explicit GitHub CLI mode authenticates as the signed-in GitHub account. Neither mode sends project files or Local AI usage reports. The check reads the manifest and signature; it does not download the release package or install an update by itself.
Your choices
- Turn the update check on or off in the installer, or with
localai policy set --update-check onand--update-check off. - Install from a release folder instead of GitHub: the installer then reads the release from the folder.
- Skip prerequisites and models in the installer and install them yourself.
- Remove expired local data with
localai prune; use--dry-runto preview the retention cleanup. This keeps current indexes of active repositories and usage reports and telemetry still within their retention period. To remove retained data as well, open the Local AI uninstaller from Apps & features and select the data categories to delete.
Changes
A change to any of this changes this page in the same pull request.
Documentation website
The interactive landing page calculates in your browser without analytics or model calls. Reference pages load Mermaid 11.12.1 from jsDelivr to render diagrams; this sends a normal web request (including your IP address and browser metadata) to that CDN. GitHub hosts the site and Wiki under its own privacy policy. These website requests are separate from the installed runtime.