What is a Knowledge Object?
A Knowledge Object is one unit of expertise made machine-readable: a single fact from an organisation's documents, research or data, stored once with when it applies, its evidence and the exact passage it comes from. It's not a document and not a chunk of text. Uthereal creates Knowledge Objects so that AI assistants and agents answer from facts, and every sentence in an answer can be traced to its source.
How is a Knowledge Object different from a chunk of text?
A chunk is a stretch of text cut from a document by length. A Knowledge Object is one fact, kept with everything needed to use it correctly. Take one clause from a supply agreement (an illustration, not a customer's contract) and one question: how much notice must we give to end this agreement?
The clause, and what a chunk keeps
14.2 Either party may terminate this Agreement by giving ninety (90) days' written notice. Where the other party is in material breach, the notice period is thirty (30) days.
Cut into chunks, the clause can split after the first sentence. A tool that retrieves only the first chunk answers 90 days and never sees the 30-day case.
The same clause as a Knowledge Object
- Statement
- Either party may end the agreement with 90 days' written notice.
- When it applies
- Ending the agreement where neither party is in breach.
- Evidence
- “…by giving ninety (90) days' written notice.”
- Source
- Supply agreement · Clause 14.2
- Linked exception
- 30 days where the other party is in material breach.
The Knowledge Object keeps the rule, its condition and its exception together, with the passage that proves each. An assistant that answers from it gives both notice periods and says when each one applies.
What is stored
A chunk of text: A stretch of text, cut to a set length.
A Knowledge Object: One fact, stored once.
When it applies
A chunk of text: Stays in the surrounding text, and is lost when the cut falls in the wrong place.
A Knowledge Object: Kept with the fact.
Evidence
A chunk of text: The chunk itself, with whatever else it happens to contain.
A Knowledge Object: The exact passage the fact comes from.
Links
A chunk of text: None. Each chunk stands alone.
A Knowledge Object: Linked to related objects, such as the exception to a rule.
For the definition in one paragraph, read the glossary entry on Knowledge Objects.
How does a document become Knowledge Objects?
The film follows one page from an archive through five steps.
- 01
From page to object
Documents are read and turned into individual Knowledge Objects.
- 02
Inside an object
The fact, when it applies, its evidence and the exact passage it comes from, in one common structure.
- 03
Everything connects
Objects link to each other and form a map of your expertise that people and AI can follow.
- 04
Trust, measured
Each object's confidence is shaped by its neighbours and its sources.
- 05
Built for machines
Every object can be reached through a clean API.
What do Knowledge Objects make possible?
Answers your people can act on, from knowledge you structure once.
An assistant or agent that answers from Knowledge Objects gives staff, members or customers the rule, when it applies and its exception together. The knowledge is created once, so each new product draws on the same facts.
Uthereal finds 95% of the facts an answer needs; typical AI tools, from assistants you can buy to RAG systems built in-house, find 59%.
Tested on four public benchmarks and published journal research, with gains on every one.
95%
Uthereal keeps misleading material out of what the AI reads 95% of the time.
Up to 5x
Up to 5x more deterministic answers: codified knowledge returns the same answer, not a guess.
Up to 4x
Up to 4x faster responses: retrieval runs on structure, with far fewer AI model calls.
Up to 10x
Up to 10x cheaper to run: agents read structure far more efficiently than raw text.
Multiples compare Knowledge Objects with vector and search applications.
Who owns Knowledge Objects, and where can you use them?
Your content and your Knowledge Objects stay 100% yours. Use them anywhere.
Create them once and use them wherever your people and customers ask questions. There are three ways to use them.
For your teams
Agents and assistants that answer from your organisation's own knowledge.
In your own AI apps
Built by you, with our API, SDK and MCP server. You keep your front end.
As a finished product
Built for you, custom-made by us, with its own interface.
They run where you choose: our cloud, your private cloud, or fully on-premise and air-gapped. For a publisher, that can be an assistant on your catalogue, under your brand, inside your site. The publishers page shows how.
Questions buyers ask about Knowledge Objects
We already have an AI assistant that shows its sources. What do Knowledge Objects add?
Many assistants show their sources. The difference is in what they find: typical AI tools, from assistants you can buy to RAG systems built in-house, read loose chunks of text and miss about 4 in 10 of the facts an answer needs. Keep what you run. Send us your toughest real questions and compare the answers side by side.
Does it hold as our content grows?
Yes. Give typical AI tools more text to read and they stop improving at about 69%; our technology keeps improving, to 98.5%. That's the share of the facts an answer needs that each one finds.
Can it run inside our own site or product?
Yes. Add it to your own app with our SDK, API or MCP server, behind your own front end. Data behind your own APIs is fetched when a question needs it. Your readers and clients stay your customers.
Is our content safe?
No AI provider keeps your content or learns from it, and your content is never used to train AI models. You choose which models and providers are used; with self-hosted open models, your content never leaves your environment. Data stays in the EU, and Swiss hosting is possible for data and for some uses with self-hosted models. SOC 2 Type II and ISO 27001 audits are underway.
Do we have to prepare our documents first?
No. Start with what you have: books, journals, guidelines and contracts, exactly as they are today. Headings, paragraphs, tables and references are parsed with over 99% extraction accuracy. You create the objects once and reuse them in every product.
How do we start?
Book 30 minutes with our team. Bring a sample of your content and the questions your experts hear most. You see your content structured into Knowledge Objects and your real questions answered, and you leave with a plan for a pilot on your own data.
Another question? Ask our team on a call.
See your own content as Knowledge Objects
Your knowledge is your moat. We make sure it stays yours.