Saved in your browser
Chats, characters, profiles, memories, documents, generated media and settings are saved in this browser using local storage and IndexedDB. Mirid reopens your last conversation and tab. This data is not uploaded to a Mirid account or synced between devices.
API keys stay in page memory by default. If you enable "Remember API keys on this device", Mirid saves them in a separate encrypted browser store. The encryption key stays in the same browser. This protects against accidentally exporting plain keys; it does not protect against someone or malicious software with access to your browser, or compromised code on this website. Keys are excluded from Mirid backup exports.
"Protect saved browser data" asks the browser to reduce automatic storage eviction. The browser may decline. Private browsing, clearing site data, storage limits or losing the device can still remove your work. Export backups in Settings.
Optional usage counts
Analytics are off until you select "Allow usage counts". Declining leaves every feature available. The choice is remembered on this browser and can be changed under Browser data or Settings.
When enabled, the app sends a small event to Mirid's own collector. Each event contains exactly three fields: an event name, a provider category and a public model ID. Events without a model use "none". Unknown or custom model names are replaced with "other" in your browser before sending.
The collector accepts only a fixed list of events and published provider model IDs. It immediately increments a daily total for that event, provider and model. It does not store a history of individual requests. The database contains only UTC day, event, provider, model and count. Daily totals are kept for 90 days in the active database; Cloudflare's database recovery facilities may retain earlier database states for their recovery period.
- Opening the browser app.
- Model selection, model requests, completed replies and generation stops.
- Speech requests and playback, image generation and transcription use.
- Document imports, character creation and backup exports, without their names or contents.
Automatic model requests, such as chat titles or enabled memory processing, count as model requests too. Counts measure activity, not unique people, and may be incomplete if a browser blocks analytics or a request fails.
What these analytics exclude
No prompts, messages, reasoning text, character information, profile information, usernames, email addresses, file names, file contents, images, audio, API keys, voice names or custom endpoint names. No conversation IDs, visitor IDs, fingerprint, cross-site identifier, page URLs or referrers. The collector does not add IP addresses, locations or browser details to the usage database.
We do not record screens, keystrokes, click coordinates or error messages. The analytics choice is a local preference, not a visitor identifier. Imported backups cannot opt you into analytics.
Providers and hosting still handle requests
When you generate a reply, image or speech, or transcribe audio, the required content goes directly from your browser to NanoGPT or OpenRouter. Selected documents, characters, conversation history and enabled memory processing can form part of those requests. Provider charges and data policies apply.
Mirid's generation requests do not pass through our analytics collector. Model catalogues contact providers. Images whose URLs you import or display can contact their image hosts. Fonts are served with Mirid instead of contacting Google Fonts.
Cloudflare hosts this website and the usage collector. Like any hosting provider, it processes network metadata such as IP addresses, request paths and times to deliver and protect the service. That also applies when usage analytics are off. Our collector's request logging is disabled, but we cannot promise that the hosting provider processes or retains no metadata.
Why we ask, and how Mirid might earn money
Mirid is independent free software. We want people to be able to use it, make it part of their daily work and help decide where it goes. Running and developing it takes time and money. We intend to build a sustainable business around it; the future pricing and paid offering are not settled.
Usage totals help us decide which models and features deserve maintenance and investment. They are not a reason to collect your conversations or build a profile of you. Your content is not part of the analytics product.
It would be misleading to call optional usage analytics "zero tracking". The narrower promise is specific: no content or visitor identifiers in these usage counts, an explicit choice before they start, and published source code you can inspect. Any expansion of the collected fields will need a new explanation and choice.
Your controls
Open Browser data in the app, or Settings > General, to turn usage counts off, change key storage or request persistent storage. Turning counts off stops future events from this browser. Already combined daily totals cannot be traced back to you or selectively removed for one visitor.
Use Settings to export a backup. Delete particular chats, documents or characters inside Mirid. Clearing this site's data in your browser removes its local data, saved keys and analytics choice; the next visit starts with analytics off.