Facts with their source
From text, tables and figures Theka extracts verifiable statements, each with the document and page it comes from.
Knowledge and wiki
Company wikis written from your documents, with sources.
Theka reads your documents, extracts the facts with document and page, groups them by topic and writes a summary page for each one that updates when the documents change. The chat uses the same knowledge to answer specific questions.
Wiki · Acme Holdingsup to date · 12 sources
Acme Holdings
Customer · contracts, SLA, contacts
Customer since March 20191, with a contract renewed for the three-year period 2024–20262. The SLA provides 99.2% availability during 8×5 hours3; there are 12 active installations across 3 plants4.
3 SLA-2024.pdf · page 6
“guaranteed availability of 99.2% per month in an 8×5 window”
How it is built
From documents to facts, from facts to topics.
Knowledge is not a generic summary: it is made of verifiable statements, each tied to the exact point in the document it comes from.
Topics
Facts are grouped by topic: a product, a customer, a supplier, a procedure. Recurring topics become pages.
Contradictions
If two documents say different things about the same topic, Theka flags it instead of picking one at random.
Safe reprocessing
When a document changes, the new knowledge replaces the old only if the reading is complete: no gaps halfway through.
Wiki
A page for every topic that matters.
For every recurring topic (a product, a customer, a supplier) Theka writes a summary page with the most reliable facts and their citations. It regenerates on its own when documents change, are moved to the bin or are restored.
- Every number in brackets opens the fact, the document and the page
- Facts from text, tables and figures, not just from text
- Comparison of facts between two versions of the same document
1contract-2019.pdf · p. 2
Contract signed on 14 March 2019 at the customer's premises.
2renewal-2024.docx · p. 1
Renewal for the three-year period 2024–2026.
3SLA-2024.pdf · p. 6
Guaranteed monthly availability of 99.2%, 8×5 hours.
4delivery-Q1-2026.xlsx · sheet 2
12 active installations across 3 plants.
To review · 1
Title: “Whole UHT milk 1L” · in the document: semi-skimmed milk
Evidence: label in the photo, page 1
ResolvedNot an error
To review
When the content contradicts the data.
While reading a document, Theka compares what it finds with the title, type and category. If they don't match, it opens a flag with the evidence: you resolve it or close it as a false alarm, and it won't bother you again until the document changes.
Shared or restricted
Knowledge respects who can see what.
Every fact inherits the permissions of the document it comes from. Anyone who cannot open the documents behind a topic cannot even see its name. For knowledge that is common to everyone, simply mark a category as “shared knowledge”.
Restricted by default
Wiki, search and chat show only the facts that come from documents the user can open.
Shared categories
Facts from shared categories (and from documents visible to everyone) feed a wiki for the whole company.
Topics for everyone
An administrator can make a topic visible to the whole organisation, leaving restricted facts where they belong.
Chat
Which renewals expire by the third quarter, and who is the contact?
Three contracts expire:
- Acme Holdings, by 14 July, contact B. Lombardinip. 3
- Studio D'Orsi, by 2 Augustp. 1
- Mossini Srl (supplier), by 22 Septemberp. 2
Chat
Specific questions, answers with the page.
For questions a wiki doesn't cover, the chat searches the documents and the knowledge, reads the right passages and answers citing document and page. Conversations are saved.
- Searches by keyword and by meaning, even inside tables
- Reads a whole document or only the pages it needs
- Sees only what the user can open: no shortcuts through AI
Tools
To keep knowledge in order.
Search by topic
A single search returns documents, related topics and individual facts with their source.
From topic to master data
A topic that is also a business object (customer, product) can be promoted to master data with its own fields.
Merging topics
Are “Acme” and “Acme Holdings” the same thing? Merge them: facts and relations are combined without losing citations.
Topic map
A graph view of topics and their links, for exploring company knowledge.
Costs under control
Knowledge works within a budget.
Extracting facts consumes tokens from your AI provider. Theka measures them and lets you decide how far to go.
Daily cap
Once the limit is reached, extraction stops; the rest of Theka carries on.
Overnight resumption
Documents left behind are picked up overnight, after the budget resets.
Image cap
A maximum number of images read per document, adjustable per type.
Estimate first
For an entire archive you see how many documents and how many tokens before you start.
Frequently asked questions
About knowledge.
Can the AI make things up?
Every sentence in the wiki and every chat answer carries a citation to document and page. If there is no source, the sentence is not written.
Who can see the wiki?
Everyone sees the wiki built on the documents they can open. Categories marked as “shared knowledge” feed a version visible to the whole company.
How much does it cost in tokens?
You set a daily cap: once the limit is reached, extraction stops and resumes overnight, while the rest of Theka keeps working. There is also a cap on the images read per document.
Can I switch knowledge off?
Yes, it is a module. Switched off, Theka remains a complete document archive; switched back on, it picks up from the facts already extracted.
Keep exploring
The other parts of Theka.
// contactLet's try Theka
Let's try Theka
on your archive.
A 30-minute call for a guided demo and an initial assessment of how well Theka fits the way you work.