Deployment
EU hosted, your own cloud, or fully on-premises. The same product in all three.
How it works
<proof>, not promisesHippocampus answers from your documents with the passage that proves it, not a plausible guess. If the material doesn't say it, it says so.
Hippocampus retrieved the right material for every answerable question.
In 3% of cases (Claude Opus 5, 93 of 96 answers correct), the connected model still answers wrong even though the right material was retrieved. That gap closes as the underlying models improve, and Hippocampus isn't locked to any one of them.
On the benchmark's unanswerable question, Hippocampus declined every time (6 of 6, Claude Opus 5) instead of guessing. It only states what the retrieved passages actually say, never inventing dates, names or numbers to fill a gap.
Insulin may be administered by a nurse, or by care staff holding a written delegation for that resident 1. Before a dose, the second-person check and the delegation reference are signed on the medication list 2.
Head to head
Hippocampus benchmarked against the top RAG services in the world. All results can be reproduced by you.
See the full resultsSecurity
Where it runs, who sees what, and what is recorded.
EU hosted, your own cloud, or fully on-premises. The same product in all three.
Each person and each agent sees only what they are cleared to see, enforced when the query runs.
Every answer records the documents that produced it.
Your material stays yours. Nothing is used to train models.
Sign up, ingest a document, ask it something hard. No credit card, EU hosted, self-hostable.