Your choice of model decides where the request goes.

This website

The Mirid browser app offers optional, content-free usage counts after an explicit choice. The hosting platform may also process operational and security data. The company hosting the site and its delivery network may keep ordinary server logs, such as an IP address, requested path, browser type and time, for operation and abuse prevention.

The browser app does not require a Mirid account. The separate desktop download access service has its own privacy notice.

Browser preview

The real Mirid frontend at /try/ saves conversations, characters, profiles, settings, imported documents and memories in this browser using local storage and IndexedDB. Refreshing does not delete them. Export a backup in Settings before clearing site data. This data is not synced to a Mirid account.

API keys stay in page memory by default. You can opt to remember them in a separate encrypted browser store so they survive reloads and browser restarts. The encryption key is stored in the same browser. Keys are excluded from backup exports. Turn remembering off under Browser data to remove the saved keys.

Chat, selected document excerpts, images, speech and transcription requests go directly from your browser to NanoGPT or OpenRouter. Enabled memory extraction can send conversation excerpts to the selected provider. These providers may retain requests and charge your account under their policies. Mirid's website server does not receive your provider keys or generation requests.

Usage analytics are off until you opt in. They collect allowlisted feature names, provider categories and public model IDs as daily totals, without content or visitor IDs. Read the complete list, storage limits and funding explanation. Model catalogues contact the provider. Fonts are served with Mirid; character images may contact the image hosts configured in your cards. Local models, native computer services and the desktop scheduler are not available in this browser build.

Data kept on your device

Mirid stores application settings, characters, conversations, story state, credentials you enter, generated media and operational logs on your computer. Local model inference does not need to send your prompt or response to Mirid.

When data leaves your device

If you choose a hosted text, image, audio, search or media provider, Mirid sends the information needed for that request to the endpoint you configured. Depending on the feature, that may include prompts, conversation context, character details, images, audio or generated instructions.

Those requests are governed by the provider's terms, retention practices and privacy policy. Mirid cannot control how an external provider handles them. Review a provider before sending sensitive material.

An OpenAI-compatible endpoint can be local or remote. Its name describes the protocol, not where the server runs.

Runtime and model downloads

On first launch, Mirid retrieves its core runtime from a public Hugging Face repository. If you use the model downloader, Mirid also contacts Hugging Face for the repositories you request. Those services receive the network information ordinarily involved in a download.

Logs and support material

Mirid writes local logs to diagnose startup and model failures. Logs can reveal file paths, provider names and operational details. Inspect them before posting or sending them to anyone, and remove credentials or personal content if present.

Remove local data

Delete conversations and characters inside Mirid when you want selective removal. Uninstalling the application may leave its data behind so an update does not erase your work. To remove the remaining Windows data, close Mirid and delete:

Deleting this folder removes the downloaded runtime and local Mirid data within it. Model files you selected elsewhere remain where you placed them.