Upload your documents
8 source types, two languages — your whole knowledge base in one place.
Research & Knowledge
Tessera·HP LIVE
AI paraphrases your own methodology back at you. Tessera answers in the author's exact words — and remembers it all between sessions.
Upload your documents and ask. Tessera finds the right passage in your bilingual knowledge base and answers with exact citations, without bending your terms. And it doubles as memory for your AI agents: Claude and Cursor plug in and see the same corpus you do.
The problem
You ask AI about your own methodology — and it answers in someone else's words.
You have a complex knowledge base in two languages. Ordinary search loses the thread between sessions and paraphrases your terms "its own way" — and won't show you where it got them. A proprietary methodology won't forgive that: one bent word and the meaning is no longer yours.
How it works
Upload it → ask it → get the citation.
Three ways to search at once
3 models + BM25 + rerank find the exact passage — not "something kind of similar."
Answer in the author's words
You see where every chunk [N] came from, and how sure the answer is — the author's words, not a paraphrase.
Memory for your AIs
Plug Tessera in as an MCP server — and every one of your AI agents works off the same corpus.
Capabilities
A knowledge base that feeds your agents.
Hybrid retrieval
3 embedding models + BM25 + rerank — precision instead of "kind of similar."
Citations and confidence
Answers with inline citations [N] and calibrated confidence.
8 source types
Documents, pages, notes — bilingual.
5 Studio skills
Brief, methodology audit, podcast-style audio overview, mind map, cross-project synthesis.
MCP server
Tessera itself is the memory for Claude, Cursor, and any AI agents.
Multi-tenant RBAC
Teams and roles — self-hosted in the EU, GDPR.
What makes it different
NotebookLM is for notes. Tessera is for methodologies.
Triple retrieval.Unlike NotebookLM — 3 embedding models + BM25 + rerank and brand-aware precision for proprietary terms.
MCP-native.Connects to any AI agent as shared memory — NotebookLM can't do that.
Self-hosted in the EU.Multi-tenant, RBAC, GDPR — your corpus never leaves for someone else's cloud.
Proof
Numbers, not promises.
$0.65
retrieval cost per year
~18.8 s
RAG answer with citations
7/7
query types pass
14×95×56
benchmark: models × chunks × queries
Offer · Value Equation
Your knowledge — answered in the author's exact words.
What you get
One memory for every AI you run
The same knowledge base feeds you and all your agents — from one place.
How we prove it
7/7 query types pass
Benchmarked across 14 models × 95 chunks × 56 queries.
How fast
~18.8 seconds
From your question to an answer with citations.
How much effort
Upload a document
Chat with citations right away — no pipelines to wire up.
No risk
Upload one document and check the citations yourself — before you move your whole corpus.
Questions
FAQ
Tessera searches three ways at once, so it doesn't garble your proprietary terms. And it doubles as memory for AI: your agents (Claude, Cursor) plug in and see the same corpus you do. NotebookLM can't do that.
It means your AI assistants pull facts from your knowledge base instead of out of thin air. Over the MCP standard, Claude, Cursor and others connect to Tessera and work off your documents — not a paraphrase from memory.
On our servers in the EU, with access split by team and role. Your corpus never goes off to train someone else's model.
Give your agents a second memory.
Bilingual. With citations. Self-hosted in the EU.