Verified against the LocalePilot source code on 30/10/2026 (commit b84b323). Ollama and LM Studio can be used without LocalePilot Pro.
What is sent
The local server receives the same content as a cloud AI model: for each text, the key name, the source text, the current translation if it is not empty, the description and group of the key, up to 6 neighboring keys of the same group and up to 2 translations of the same key in other languages. For the whole request: the language codes, the approved glossary terms if Use approved glossary is on (Pro), and the project's Translation context if you filled it in. The details are on What OpenAI, Anthropic, Gemini and DeepSeek receive.
Adapt with AI, on the Store listings screen, also works with a local model.
The AI audit also works with a local model and sends the same content, for already translated cells of the audited language only. It requires LocalePilot Pro, even with Ollama or LM Studio. See AI audit.
What is never sent
- Your files, their names and their paths.
- The name of your project, except where it appears inside a text you translate.
- Forbidden glossary terms and glossary terms that are not approved.
Does it leave your Mac?
| Provider | Default address | API key |
|---|---|---|
| Ollama | http://localhost:11434 | None |
| LM Studio | http://localhost:1234/v1 | Optional, sent in an Authorization: Bearer header |
- With the default address and a model downloaded to this Mac, the request goes to a server on this Mac and nothing leaves it. A cloud model offered by Ollama or LM Studio receives the text even at the default address.
- The Base URL can be changed in Settings, AI provider, for example to use a server on another computer. The text then leaves your Mac and goes to that address.
- Ollama and LM Studio can themselves be set up to use a remote service. That depends on their own configuration, not on LocalePilot.
The Before you launch panel, in the AI batch translation window, shows the real address and whether it is local, on your local network, remote or an unknown destination. LocalePilot warns you when an http:// address does not point to this Mac, because the text would travel unencrypted.
What stays with the provider
LocalePilot creates nothing in Ollama or LM Studio: it only uses the models you have already downloaded. To detect them, it reads the list of models from the local server and, for Ollama, runs the ollama command on your Mac.
API key
Ollama has no API key. For LM Studio, a key is optional: add one in Settings, AI provider if your server requires it. It is stored in the macOS Keychain, never in LocalePilot's settings file.
Options that change what is sent
The same as for the cloud AI models: Use approved glossary, the project's Translation context, the Batch size and Translate automatically with … in Add a language. See Options that change what is sent.