Privacy Policy

Last updated: October 30, 2026

Privacy at a glance

This policy covers LocalLM Lab, and JevDK and jev-serve (see JevDK and jev-serve). LocalLM Lab runs on your own Mac. By default, it uses Apple's on-device Foundation Models framework (Apple Intelligence) to generate responses directly on your Apple Silicon hardware, with nothing leaving your device. In AI Models, you can also choose to route a request to Apple's Private Cloud Compute, or to a hosted online provider you configure yourself — OpenAI, Anthropic Claude, OpenRouter, or any other OpenAI API-compatible service — in that case, your prompt and its response travel directly between your Mac and that provider, using your own credentials, and are governed by that provider's own privacy practices, not this policy. Whichever model you use, LocalLM Lab's developer, ThisBrainAI LLC, does not receive your prompts, responses, or any content you enter into the app. There is no cloud backend, no account, and no server operated by ThisBrainAI LLC that your data passes through — LocalLM Lab has nothing to send it to even if it tried.

Information processed on your device

When you use Prompt Playground, your system prompt and input are sent directly to whichever model you've selected in AI Models. With Apple's on-device model or a downloaded open-weight model, nothing you type leaves your Mac. With Private Cloud Compute or a hosted online provider (including Claude), your prompt and input are sent from your Mac directly to that provider — never through ThisBrainAI LLC.

When you use API Lab, LocalLM Lab runs a local, OpenAI-compatible API server on your own machine so other apps or scripts you run can talk to whichever model is selected. By default this server listens on localhost only. If you choose to expose it to your local network, requests still stay on your LAN and are never routed through ThisBrainAI LLC's infrastructure, because ThisBrainAI LLC does not operate any infrastructure that LocalLM Lab talks to.

Because generation with Apple's on-device model or a downloaded open-weight model happens entirely on your Mac, your prompts and outputs in that case never reach any third party at all. Using Private Cloud Compute or a hosted online provider instead means your prompts and outputs are subject to that provider's own processing behavior — see the next section.

AI Models and third-party providers

The AI Models panel lets you add Apple's Private Cloud Compute or a hosted online provider — OpenAI, Anthropic Claude, OpenRouter, or any other OpenAI API-compatible service — each using your own account or API key. LocalLM Lab stores each provider's key locally on your Mac, in the macOS Keychain, and never transmits it to ThisBrainAI LLC. Selecting one of these models sends your prompt, your input, and any connector or MCP tool data included in that turn directly from your Mac to that provider — not through ThisBrainAI LLC, because ThisBrainAI LLC operates no infrastructure for LocalLM Lab to route through. Once your data reaches that provider, it is handled under that provider's own privacy policy and terms, which you should review separately before using it.

Open-weight models you download (from Hugging Face) run entirely on your Mac, the same as Apple's on-device model — no network call is made at inference time once the model is downloaded.

API Lab access tokens

If you enable API Lab, LocalLM Lab generates a local access token used to authenticate requests to your local server. This token is stored locally on your Mac and is never transmitted to ThisBrainAI LLC. Regenerating the token immediately invalidates the previous one.

Request and response logging

Prompt Playground logs every run to a plain-text file on your Mac at ~/Library/Logs/LocalLM Lab/dev.log, including the system prompt, user input, each tool call with its arguments and output (which can include data from connectors and MCP servers, capped at 20,000 characters per output by default), and the model's answer. This happens by default, the file is not cleared automatically, and it is not sent to ThisBrainAI LLC.

If you enable request logging in API Lab, each incoming request and response is appended to a plain-text file on your Mac at ~/Library/Application Support/LocalLM Lab/api-requests.log. This is off by default, only writes when you turn it on, and is never sent to ThisBrainAI LLC.

LocalLM Lab also writes diagnostics from its background helper (the process that runs the models) to a plain-text file on your Mac at ~/Library/Logs/LocalLM Lab/helper.log. It records things like model repository names, model pairing and version-pin activity, provider errors and crashes. It is not intended to contain your prompts or the model's outputs, which are recorded only in the logs above. The file is readable only by your user account, is kept to about 1 MB plus one older copy, and is never sent to ThisBrainAI LLC.

These log files stay on your device and can be deleted at any time by removing the files directly. (When API Lab request logging is on, its settings screen has a Show in Finder button for the request log.)

Information stored locally

LocalLM Lab stores non-secret settings locally on your computer, such as your chosen prompts, server configuration, per-model tuning settings, the exact Hugging Face version each downloaded model is pinned to, and preferences. This data stays on your Mac and is not transmitted anywhere by LocalLM Lab.

Telemetry

LocalLM Lab does not collect usage analytics or telemetry.

Deleting local data

You can remove LocalLM Lab's local settings, model version pins, and log files by deleting the folders ~/Library/Application Support/LocalLM Lab and ~/Library/Logs/LocalLM Lab (uninstalling the app does not remove them), and its saved API keys and tokens by deleting the LocalLM Lab items in Keychain Access.

Mac App Store version

The Mac App Store version of LocalLM Lab works the same way as described above, with these differences. It has no API Lab, so it runs no local server and creates no API Lab token. Before it first sends anything to an online service — a hosted model provider, Claude, or an MCP server — it asks for your permission, naming the service and the kind of data it will receive; the services you've allowed are listed in its Security panel, where you can revoke them. Because it runs in Apple's App Sandbox, its settings, logs and downloaded models are kept in its own container (~/Library/Containers/ai.thisbrain.locallmlab/) rather than the locations listed above, and its API keys and tokens are separate Keychain items from those of the downloadable version. It requires macOS 27 or later.

Requirements

LocalLM Lab requires a Mac with Apple Silicon, macOS 26 or later, and Apple Intelligence enabled; Private Cloud Compute, open-weight (MLX) models, and hosted online providers need macOS 27. Apple's own terms and privacy practices govern how the Foundation Models framework processes data on your device, and how Private Cloud Compute processes it on Apple's own servers; please refer to Apple's documentation for details. If you use a hosted online provider, that provider's own terms and privacy practices govern how it processes your data once it leaves your device.

JevDK and jev-serve

This policy also covers JevDK, ThisBrainAI LLC's app for designing and testing decision models, and jev-serve, the command-line server that comes with the downloadable version of JevDK. They work the same way as LocalLM Lab: no account, no analytics or telemetry, and no server operated by ThisBrainAI LLC. ThisBrainAI LLC does not receive your questions, inputs, marked answers, results or any other content you enter.

On your Mac. Your questions, inputs, marked answers, calibration and results stay on your Mac. JevDK keeps your current workspace between launches in its own settings, and writes workspaces, exported questions, server configs and CSV files only to files you choose. With a local model, everything JevDK asks the model happens on your Mac.

Hugging Face. Models are downloaded from Hugging Face. Checking or downloading a model contacts Hugging Face for that model, and searching in Models sends your search text to Hugging Face; once a model is downloaded, no network call is made to use it.

Hosted backends. JevDK can also ask hosted decision models: Featherless, TypeSafe's Jev through OpenRouter, and OpenAI's Decisions API (OpenAI), each with your own API key. Hosted backends are off until you turn one on. While one is on, each input you run and your questions (their text, and the answers you allow with their descriptions) are sent from your Mac directly to that service, never through ThisBrainAI LLC, and are handled under that service's own privacy policy and terms. Your system instructions are not sent. To stop sending (and withdraw your permission), turn the backend off in Backends; nothing more is sent to it from then on. To have data already sent deleted, contact that service, since JevDK and ThisBrainAI LLC keep no copy of it. The downloadable version also offers Featherless's free demo service, which needs no key. API keys are stored in the macOS Keychain and are never sent to ThisBrainAI LLC.

jev-serve. jev-serve answers requests on your Mac, by default only from your Mac itself (127.0.0.1). It writes a one-line summary of each request to its own terminal output, and the request and response bodies only if you start it with --log-bodies; it doesn't write log files. A server config exported from JevDK can hold an access token; the file is readable only by your user account.

Mac App Store version of JevDK. It works the same way, with these differences: it doesn't include jev-serve or Featherless's free demo; it asks for your permission each time you turn on a hosted backend, naming the service and what it will receive; and because it runs in Apple's App Sandbox, its settings and downloaded models are kept in its own container (~/Library/Containers/ai.thisbrain.jevdk/), and its API keys are separate Keychain items from those of the downloadable version.

Deleting JevDK's data. For the downloadable version, delete ~/Library/Preferences/lab.locallm.sdk.reference.jevdk.plist (settings and the saved workspace), ~/Library/Application Support/jev-serve (jev-serve's model version pins), and the JevDK items in Keychain Access. Its models are in the shared Hugging Face folder, ~/.cache/huggingface/hub, which other tools may use too; remove a model there, or in JevDK's Models panel. For the Mac App Store version, delete the app, its container folder (~/Library/Containers/ai.thisbrain.jevdk), and its ai.thisbrain.jevdk items in Keychain Access.

Contact

For privacy or support questions, contact neuron@thisbrain.ai.