Why troveGEN
Your AI is only as good as the passages it reads.
Ask an AI model about your own documents and it can only answer from what it is handed. Hand it the wrong passage and it answers wrongly, with confidence. troveGEN is the layer that finds the right passage, from documents of any kind, for only the people allowed to see it.
New to this?
What is RAG, and why does it matter?
In plain words
AI models know what they were trained on. They do not know your contracts, policies, price lists or clinical guidelines. RAG (retrieval-augmented generation) closes that gap in two steps: first find the passages in your documents that matter to the question, then hand only those to the AI.
Everything depends on the first step. If search misses the answer, reads a table as a jumble of words, or returns something the user should not see, no AI model can recover. That first step is what troveGEN does. A longer explanation →
The problem
Three things go wrong between your documents and a good answer.
Most teams meet all three, usually in this order, usually after the demo worked.
Documents are messy
Real documents are not clean text.
- A scanned page returns nothing, and nobody notices.
- A table turns into a jumble of words, so its numbers cannot be trusted.
- A clause is separated from the sub-clause that changes its meaning.
Search misses, or invents
The AI can only use what search hands it.
- The right passage ranks fifth, so it never reaches the AI.
- The same question, worded differently, finds nothing.
- When nothing matches, a model answers anyway, and sounds sure.
Access is an afterthought
Everything lands in one searchable pile.
- Staff-only and customer documents end up side by side.
- Account numbers and personal details sit in a search index.
- A deleted document lives on in copies nobody remembered.
What you get
Six answers to the questions every careful buyer asks.
Each one says what you get first, then how it works.
WILL IT FIND THE RIGHT PASSAGE?
It reads structure, then searches two ways at once.
Scanned pages are read, tables stay tables, and headings travel with every passage. Exact words and meaning are searched together, then ranked. Built-in evaluation tools score the results on your own documents, so a change is proven before you ship it.
WILL IT MAKE THINGS UP?
Every claim is cited, and it says "not found" when it should.
Cited answers are optional. When you turn them on, each citation is checked against a real passage, questions your documents cannot answer are refused, and English answers are scored for how well their sources support them.
WHO CAN SEE WHAT?
Permissions are part of the search itself.
Access rules are applied inside the query, before ranking, so a restricted document cannot leak through a cleverly worded question. Account numbers, national IDs and similar details are replaced with tokens before anything is indexed.
WHERE DOES OUR DATA GO?
Wherever you decide.
Hosted for you in a dedicated space, or written into your own vector database, or run entirely inside your network with no outside calls. troveGEN does not train any model on your documents. Deleting a document removes what was derived from it and gives you a receipt.
ARE WE LOCKED IN?
No. Every boundary is a connection you own.
Use Qdrant, Pinecone, pgvector or plain Postgres, and your own embedding and language models, or ours. It is a plain REST API with a typed SDK. Moving between hosted and self-run means loading the documents again into the new home.
WHAT WILL IT COST?
You pay per page read, and you can start free.
The free trial is a one-time budget of 500 pages and 2,000 searches, enough to judge quality on a real corpus. Billing is per page, not per document, because a one-page note and a 400-page manual are not the same work. Models you bring yourself cost nothing extra here.
Who it is for
Built for two kinds of team.
The product is the same. What each team needs from it first is different.
AI STARTUPS AND PRODUCT TEAMS
Ship document search this week, not this year.
- One API. A typed SDK and an MCP server let your app or AI agent search your documents without you building the plumbing.
- Nothing to run. Start free with 500 pages. No vector database to host and no model to deploy.
- Your choice of model. Works with whichever language model you already use.
- Proof for your customers. Built-in evaluation shows how well search works on their documents.
- A path upward. When an enterprise customer asks for self-hosting, the same product runs in their environment.
CORPORATES AND REGULATED TEAMS
Control, governance and proof for the security review.
- Your infrastructure. Your database, models and keys, or fully air-gapped with no outside calls.
- Access in every search. Permissions are enforced inside the query, and changes are recorded in an audit trail.
- Sensitive data protected first. Replaced with tokens before indexing; deletion removes derived data and gives a receipt.
- Tuned for your documents. Finance, healthcare, legal, hospitality and government material is handled the way each industry needs.
- Roles that match your org. Admins, developers and users, with project-level access for each person.
Industries we serve
Documents are different in every industry.
troveGEN recognises what kind of document it is reading and handles it the way that industry needs, so answers come back complete and correct.
Finance
Statements you can search and add up. Filings that stay in one piece.
Learn more →Healthcare
Clinical lists come back complete, never a fragment.
Learn more →Legal
A clause stays with its sub-clauses. A judgment stays in context.
Learn more →Hospitality
Policies, rates and menus answered accurately, every time.
Learn more →Government
Every circular found by its reference number, date and department.
Learn more →Any other document
Text or scanned, structured or messy. If it holds words, troveGEN can search it.
Learn more →Deploy your way
Same protections, wherever it runs.
You choose where the data and the models live. The safeguards come with every option.
| We host it (troveGEN-Z) | Your vector database | Sovereign (your environment) | |
|---|---|---|---|
| Who runs the vector database | Us, in a dedicated space | You | You |
| Embedding and language models | Ours, or yours | Yours, or ours | Yours, on your servers |
| Permissions enforced inside every search | ✓ | ✓ | ✓ |
| Sensitive details tokenised before indexing | ✓ | ✓ | ✓ |
| Deletion removes derived data, with a receipt | ✓ | ✓ | ✓ |
| Built-in evaluation of search quality | ✓ | ✓ | ✓ |
| Runs fully air-gapped | — | — | ✓ |
| Starts at | $25 per month | $89 per month | Quoted |
Questions we hear first
Straight answers.
Is troveGEN a vector database?
No. A vector database stores and searches vectors. troveGEN is everything around it: reading documents, keeping their structure, searching, ranking, permissions and evaluation. It writes into the vector database you already chose, or hosts one for you if you would rather not run one.
Do we need to understand RAG to use it?
No. You upload documents, call one search endpoint and get back passages with their sources. Add AI-written answers when you want them. The console also lets your colleagues chat with the projects you choose.
Can we use our own AI model?
Yes. Any model that speaks the OpenAI-compatible interface works, including models on your own servers. Answers are off by default, so nothing needs a language model until you ask for it.
How quickly can we start?
A first search over a handful of documents takes minutes. A 400-page scanned book takes longer, because every page has to be read before it can be searched.
What does troveGEN not do yet?
We hold no third-party compliance certification yet. Chinese, Japanese, Korean and Thai documents are not accepted yet, and eighteen other languages are verified end to end. It is a retrieval layer for your product, not a ready-made chatbot for your customers. We would rather say so now than have you find out later.
See it on your own documents.
The free trial is a one-time budget of 500 pages and 2,000 searches. No card needed. If you would rather talk it through first, we are happy to.