mozilladatacollective.com

The Mozilla Data Collective is a catalog of ethically sourced AI training datasets — speech, text and other modalities — contributed by community organizations, each setting its own license and access terms. This server is a read-only discovery surface over the public catalog. Use it to find datasets and describe what exists. It cannot download data, read private datasets, or make purchases. Recommended workflow: 1. Call `search` with the user's need phrased as natural language. The backend runs a hybrid lexical and semantic pipeline, so a descriptive phrase ("conversational Swahili audio for ASR fine-tuning") retrieves better than a bare keyword. 2. To narrow results by task, language, license, format, price or sample availability, call `list_filters` first and use the exact values it returns. Filter values are matched exactly and are case-sensitive, so never guess or invent them. 3. Call `fetch` with an id from the search results to get the full description, license, size and pricing for one dataset. 4. Give the user the dataset's `url`. Accepting the contributor's terms, payment and downloading all happen on that page. For programmatic access, point them to the REST API and Python SDK documented at /api-reference. Ground every answer in what the tools returned: only name datasets that appeared in a result, and quote license and pricing from `fetch` rather than inferring them. If a search returns nothing, say so plainly — the catalog does not cover every language or task, and a report of "no matches" is more useful than a near miss presented as a match.