Privacy Policy

Studio runs on our servers, so it handles more than the library does. In plain terms:

• What you type. Your description or schema is sent to a language model provider that runs on our instructions, to design the tables and to review the result. The model also receives summaries of the generated data, such as ranges and category shares. It does not receive the generated rows.
• Web research. By default Studio searches the public web with queries written from your description, to ground the data in real facts, and may look up public datasets on Kaggle by topic to calibrate shapes. The queries go to the search providers named below; your schema and generated data do not. You can switch research off for a dataset in the composer.
• What we store. Your account, your credits, and the recipe of each dataset (your request and the settings that rebuild it). Generated files are kept on our server for your account and are temporary. Where archiving is switched on, a finished dataset is also stored as one file in a private storage bucket for up to 14 days, so it survives a server restart. The recipe can rebuild any expired dataset. You can delete any dataset and its files and archive are removed.
• Your own model key. If you add one, it is encrypted at rest and bound to your account and that provider.

More detail, including what we do not have yet, is on the Your data and security page.

The Python package and the MCP server: nothing. There is no analytics, no telemetry and no phone-home of any kind in the library. Installing it and running it produces no network traffic to us, and we have no way of knowing you are using it.

Three things reach the network, and each is something you ask for:

• Seeding a database. The seed_database tool connects to the database you point it at, using the connection string you supply. That connection is between your machine and your database. We never see the connection string or the data.
• Vocabulary enrichment. If you enable it, Misata fetches public reference lists from Wikidata to make generated values more realistic. It sends a category name and nothing about your schema or data.
• AI schema design. If you supply your own API key for an LLM provider, the text of your prompt is sent to that provider under your own account and their terms. This is off by default. No key, no request.

The site uses privacy-preserving analytics to count page views and a small number of product events (a dataset downloaded, a generation started, finished or failed). These record what happened, not who did it. We do not build advertising profiles and we do not sell anything to anyone.

Cookies. Signing in sets one cookie, from Clerk, that keeps you signed in. It is strictly necessary and is set regardless of any choice below. Vercel Analytics, which we use for page views, sets none. Google Analytics is the one cookie-based tool on the site, and it only loads if you click Accept on the banner; declining, or never choosing, means it never loads at all. That choice lives in your browser and you can change it any time by clearing this site’s local storage.

If you create an account, authentication is handled by Clerk, which stores your email address and login credentials on our behalf. If you send us a message through the contact or ideas form, we keep what you wrote and, if you gave one, your email address, so we can reply.

Email. With an account, we keep your name, email address and when you were last signed in, read from Clerk, so we can write to you: a welcome when you join, a note when a build finishes after you have left the page, and now and then a letter about what we are building. Letters can mention what you made here (how many datasets, the latest one’s name, your credit balance), and every one has a one-click unsubscribe. You can switch off build notes and letters in Settings, under Email. We keep a record of each mail sent (address, subject, when, whether it was delivered) so nobody gets the same one twice.

If you use the public /try page without signing in, the schema you submit is generated and streamed straight back to you. It is not stored.

The website runs on Vercel and the API on Railway, with a Neon database for recipes and account records. Account authentication uses Clerk, and email is sent through Resend. Payments, when they are switched on, are handled by Polar as merchant of record; we never see or store card details.

For Studio specifically: your description goes to the language model provider we use (IFM by default, or Google Vertex AI), or to the provider of your own key if you add one. Research queries go to a web search provider (Tavily, Brave, DuckDuckGo or Wikipedia, whichever is available) and topic searches to Kaggle. Where archiving is on, finished datasets are stored in Supabase Storage. Where error reporting is on, crash reports go to Sentry and may include the request that failed.

These providers process data on our instructions in order to run the service.

We do not sell personal data, and we do not share it with anyone except the providers above and where the law requires it.

Analytics events are aggregated and retained for up to 24 months. Messages you send us are kept until you ask us to delete them. Account records are kept while the account exists and removed within 30 days of deletion. Generated data from the public try page is never written to disk.

You can ask for a copy of what we hold about you, ask us to correct it, or ask us to delete it, by writing to the address below. If you are in the UK or EU, you also have the right to complain to your data protection authority.

The most complete choice is available to everyone: use the open source package. It runs entirely on your machine and reports nothing.

Misata is not directed at children under 16. We do not knowingly collect information from them.

If what the software does changes, this page changes first, and the date at the top moves. Material changes to how personal data is handled will be announced on the site.

Misata is built and operated by Muhammed Rasin. Questions, requests, or corrections: hello@misata.studio.