Knowledge & retrieval
Answers from your own knowledge, with the source attached.
A governed knowledge base, retrieval that respects your permissions, and an assistant over the top. Built from open source, integrated with what you already run, owned by you.
- Answers
- Cited, permission-aware
- Sources
- 27 connectors + custom
- Languages
- Ask in one, answer from another
- Accuracy
- Measured on your questions
In short
An on-premise knowledge assistant answers employees’ questions from your own documents and systems, citing the source of every answer and only using content each person is already allowed to see. Retrieval, models and logs all run on your infrastructure.
What you get
More than search.
Permission-aware retrieval
Answers draw only on content the person can already open in the source system.
Cited answers
Every answer links to the document, page or record it came from.
Governed knowledge base
Document owners, review dates and checks that flag outdated content.
Many languages
Answers in the language people ask in, even from documents written in another.
Fine-tuned small models
Where they beat retrieval alone, trained on your domain with the weights yours.
Evaluation you keep
Accuracy measured against a test set built from your own questions.
How it works
From scattered sources to one trusted answer.
Ingest
Connectors and document ingestion read files, pages, tickets and records, including scanned documents.
Index
Content is embedded and indexed in a vector store on your servers, with permissions attached.
Retrieve and rerank
The best passages for each question are found and reranked under the asker’s access rights.
Answer and cite
A model you choose writes the answer from those passages and cites them; the exchange is logged.
Kept accurate
Knowledge that doesn’t go stale.
- Every document has an owner and a review date.
- Checks flag content that is outdated or contradicts newer sources.
- Owners are asked to confirm or fix flagged content.
- Answers prefer recently reviewed sources.
FAQ
Questions, answered.
How accurate are the answers?
Every answer cites its source. In the proof of concept we measure accuracy against an evaluation set built from your own questions, and that set stays with you.
Does it respect our existing permissions?
Yes. Connectors carry each source’s permissions, and roles come from your existing directory.
Which tools do you build on?
Typically Onyx, Haystack, LlamaIndex or LibreChat for the assistant layer, with pgvector, Qdrant or OpenSearch as the index. The mapping step decides.
Related
Connectors
Drive, SharePoint, Notion, Confluence and more.
Learn more →Access control
Who uses which agent, on what data.
Learn more →Fine-tuned small models
When a small model beats retrieval alone.
Learn more →Start with two weeks of evidence, not a sales call.
A fixed-price discovery sprint, credited against whatever comes next. Or write to sales@deepvox.ai.