Dataset policy
How releases are built, licensed and versioned.
Free access
Contributed conversations are free to read and download. We don’t sell or paywall them.
Licenses
- Community contributions: CC BY 4.0, attributed to the contributor’s chosen name, pseudonym or “anonymous”.
- Imported datasets: keep their original license and attribution, and ship in their own release partition.
- The The Chat Archive software license (Apache-2.0) does not apply to the data.
- AI providers’ terms may restrict how model outputs can be used. Researchers are responsible for checking them.
External datasets
Being on Hugging Face doesn’t mean a dataset may be republished. We mirror content only from datasets whose licenses clearly allow redistribution, apply our own privacy filtering, and list everything else as catalog-only entries with a link to the source. We never bypass gated access.
Releases
Releases are immutable snapshots of approved, non-suppressed records, with JSONL.gz and Parquet files, SHA-256 checksums, a manifest, methodology and known limitations. Development fixtures are never included.
Representativeness
The archive is self-selected and does not represent all AI users or uses.