The STARGATE release, searchable.

Why this exists

The release is public but hard to work with. CREST, the CIA’s own archive, holds every document, but its transcriptions are poor and its site is hard to navigate. This tool puts a new machine transcription of all12,186 documents behind one search box, each result keyed to its CIA document ID.

How to search

• Several words find documents that contain all of them. Typing “and” between them is optional. “Or” and “not” are not supported.
• Put a phrase in quotation marks to find those words together. One quoted phrase per search works best.
• Machine reading misreads names. If a name you expect is missing, try a variant: a letter dropped or doubled, or two words run together. A search that finds nothing is not proof the record is silent.
• Every result carries the document’s CIA ID. Use it to find the original scan on CREST.
• Not sure how a name is spelled in the files? The words page lists close spellings, acronyms and common phrases, with how many documents use each.
• Not sure what a code name means? The glossary defines the program’s names and terms in the documents’ own words.
• Rather wander than search? Browse every document in CIA ID order.

Download the text

The full text is free to download from Hugging Face, for use in your own search, LLM or RAG setup. It comes two ways: one record per page, and passages of about 350 words ready to embed. Each record carries its CIA document ID, page number and a link to the scan. The government text is public domain; our structure and notes are CC BY 4.0.

A companion tool

For meaning-based search across millions of declassified U.S. government documents, well beyond this collection, try HistoryLab AI.