How Historia uses a collection, in plain terms. Each line below describes what the code does today.
When a student asks a question, Historia’s server sends keywords to the public search API or catalogue search of each archive that fits the question, and reads what comes back: titles, creators, dates, descriptions and links.
To check a quotation, or to say which page of a long item to read, it may also fetch the public text or file the archive publishes for that one item, such as OCR text or a transcription. It does not crawl a site or follow links from page to page.
Link only. The student is sent to the archive’s own page for the item. Historia does not host, embed or republish images or files. The one exception is archive film from the Library of Congress, the National Archives and the Internet Archive, which plays in the page from the archive’s own servers, and only after the student presses play.
Archives searched today: Library of Congress · Chronicling America · Smithsonian Institution · Digital Public Library of America · Internet Archive · OpenAlex · Semantic Scholar · Europe PMC · ERIC · National Archives Catalog (NARA) · The National Archives (UK) · Gallica · Wellcome Collection · The Metropolitan Museum of Art · Art Institute of Chicago · Old Bailey Online · GovInfo · CourtListener · legislation.gov.uk · World Digital Library · DigitalNZ · Wikisource · Hansard · Deutsches Textarchiv
Each result shows the archive’s name and the repository the archive’s own record gives. The title, creator, date and link are copied from the archive’s response. The AI never writes them.
A quotation is copied from the archive’s own text by its character position and checked again on the server before it is shown. It always appears with its record and a link to it.
A citation Historia formats for a student names the repository and gives the record’s link.
So that a repeated search does not ask an archive again, Historia keeps four things for seven days: an archive’s answer to a search (the fields it returned and any text that came with them), the finished answer to a question (titles, links and short quotations), whether a record’s link still opens, and the page pointers for an item. After seven days the next search asks the archive again.
No image or file is stored. A file read for page pointers is read and discarded; only page numbers, match counts and a short line of text around each match are kept.
To ask for a purge, write to thehistoria.company@gmail.com and name your service. We delete those copies from the database by hand.
Requests from Historia’s servers name Historia and a contact address in the User-Agent header. Searches send Historia/1.0 (historia.organization@gmail.com). The page-pointer lookup sends Historia/1.0 (thehistoria.company@gmail.com) where-to-look.
A search sends a few requests to each archive that fits its question, at the same time. General searches of loc.gov are capped at four per search. Identical requests running at once are sent once and shared, and a search identical to one already running waits for it instead of asking again. Semantic Scholar is asked at most once every 1.1 seconds from each server.
No request waits longer than 60 seconds. A refused request is not retried, except Semantic Scholar’s within the same search. An error from your service is logged with its host name and status code only.
Write to thehistoria.company@gmail.com, ideally from your institution’s address. We can stop sending requests to your service, ask less often, change how results from you are labelled or credited, or clear what is cached from you. Each is a change we make by hand, and we reply to say when it is done.
If you agree, Historia shows a small, labelled strip, “Teaching resources from partners”, below the archive results when a student’s search matches topics you choose. It carries your credit line as you give it and a link that opens your own site.
It is a link only: nothing of yours is copied or embedded. It never appears among the archive records or in place of them, shows at most three links, and does not appear when nothing matches.
We add it only after a written yes, and remove it when you ask. Approved partners are listed on the press page.
Write to thehistoria.company@gmail.com. A person replies.