AI Potluck
Contribute
Back to Gap Map Model components / Data hubs & catalogs

ATLAS

AI Singapore
open source / Overall score: n/a

ATLAS is AI Singapore's catalog of datasets for underrepresented languages, centered on Southeast Asia. It describes each dataset by language, modality, license and task, imports dataset cards from Hugging Face, and publishes contributions after a human review. A chat search and a read-only MCP API let people and agents query the catalog, and coverage maps show which languages still lack data.

Scored as the hosted catalog at atlas-data.ai. Its privacy policy calls it open source while its terms of use reserve the site's source code, and the openness reading follows the privacy policy until AI Singapore says which is meant.

Openness

5 low confidence
5.0
license
unnamed open-source license(the privacy policy calls the platform open source
core features withheld
no — run by AI Singapore, a national research programme, with free accounts

AI Singapore's privacy policy describes ATLAS-Data as an open-source data platform, and a national research programme runs it as a free public service. No repository or license for it has been published, and its terms of use reserve the site's contents including source code, so this reading is provisional until the code appears or AI Singapore explains the statement.

Adoption

not assessed

ATLAS publishes the size and language coverage of its catalog but no count of searches, downloads or users. Catalog size describes what is listed rather than how much it is used, so no reading was recorded.

  • https://atlas-data.ai/en recorded 2026-10-08

    Home page: browse, chat and MCP routes into the catalogue and a coverage map; no usage or download counter.

Capability

1 medium confidence
1.0

ATLAS describes datasets that mostly stay where their publishers put them, such as Hugging Face, and its licensing policy says the site does not itself distribute dataset content unless a listing says otherwise. Its team can upload a new dataset for a contributor, but no interface serves datasets by identifier, so it sits with Masader rather than with UCI, which holds what it lists.

  • https://www.atlas-data.ai/en/data-licensing-policy recorded 2026-10-08

    "Unless expressly stated otherwise, the Site does not itself provide, distribute or grant rights to download or use the underlying Dataset content."

  • https://www.atlas-data.ai/en/docs/how-to-contribute recorded 2026-10-08

    "Paste a HuggingFace URL" or "Have our team upload it" ("Direct browser uploads are paused"); then "Human review" and "Published ... with a unique dataset ID".

  • https://www.atlas-data.ai/en/docs/verification recorded 2026-10-08

    Account tiers: "Signed in: A free account"; signed-in accounts "download datasets that allow it"; a Contributor "can add data to the catalogue", granted on request.

  • https://www.atlas-data.ai/en/mcp recorded 2026-10-08

    MCP tools: Dataset Search, Catalog Facets, Dataset Health, Config Listing, Schema Inspection; "The API can only read data — nothing can be modified or deleted."

Verified 2026-10-08, except some scores, still awaiting re-confirmation