Papers, scans, notes and the mail with co-authors become one collection with a citation for every claim. Semantic search works across languages; the evidence pack assembles an answer within a token budget and says what it could not find. Your library stays your library.
An illustration of the shape of an answer. The files are made up; the behavior is docenta's.
BEIR SciFact nDCG@10 of 0.72 with the dense channel and 0.66 lexical, within a point of the published baselines; LongMemEval evidence recall@5 of 0.97. The rigs are in the repository.
The evidence pack carries the route it took, the cited spans, the conflicts between sources and the questions the collection cannot answer. A PDF whose text reads out of order is flagged, never quoted blind.
Dense retrieval finds the German paper from an English question. Entity timelines put a concept's history in order across papers, notes and mail.
docenta is retrieval, not a model. Pair it with the model you already use; it hands over the evidence and the receipts.
Back to the collection · also for lawyers, insurance, home, mail, developers, researchers
docenta is a commercial product of SKY, LLC, now in private preview. Tell us what your agent needs to remember and we will tell you when the door opens.
request early access or check back here: the public beta will be announced on docenta.ai