ABOUT THE ARCHIVE

Preserve the exchange.
Keep its context.

AIRef turns a shared AI conversation into a frozen, citable record. It stores its own copy of the approved transcript, rather than depending on the original share link to stay available.

What is preserved?

The first version preserves text and code, the original source URL, user-provided reference details, capture information, and the time AIRef archived the record. Each record has a stable UUID URL and downloadable JSON, Markdown, BibTeX, and an integrity manifest.

Files, images, audio, generated artifacts, and hidden provider instructions are outside this text-only archive. A browser-assisted capture may contain only the messages currently loaded on the page. Capture scope and known omissions appear on each record.

Review before you publish.

Drafts are temporary and private to your browser tab. Automatic retrieval sends the supplied public URL to its provider. Only the reviewed transcript and supplied reference details are stored when you publish. Invitation codes are not included in records, and AIRef does not request AI-provider passwords or API keys.

Remove confidential material and personal information before publishing, including from the title and reference details. Pattern checks can flag some secrets; they cannot guarantee that a transcript is safe to publish. Only publish content you have permission to share.

Integrity has a precise meaning.

The SHA-256 manifest lets you check that the downloaded JSON matches the archived bytes. It does not prove that the provider generated the text, that the conversation is complete, or that its contents are correct. Names, affiliations, and ORCID iDs are user-provided and have not been identity-verified.

Capture details distinguish a receipt for a server retrieval from a browser or JSON import. A server receipt covers the retrieved source bytes; transcript extraction and review take place in the browser and are not independently authenticated.

Stable URLs, with practical limits.

Record URLs do not expire, and published transcript content is not silently overwritten. Corrections create a new record. Availability depends on maintaining the domain and hosting account, and remains subject to free-hosting limits. The service may pause new submissions when capacity is reached; existing published records are not deleted to make space.

Withdrawal and contact

To report sensitive, infringing, or otherwise inappropriate material, contact the archive operator through ojwatson.co.uk and include the record URL. A withdrawn record keeps a citation page while its transcript and downloads become unavailable. Withdrawal does not recall copies others have already downloaded.

Archive a conversation →