Available4-week evaluation on one bounded domain
Give AI agents reliable knowledge when they're trusted with enterprise work. Amorfs Knowledge Packs deliver exact records, with identity, relationships, time and source already attached.
20-minute demo → four-week evaluation → report scorecard
The challenge
Companies want agents that can answer, explain and act on validated internal data.
Questions like:
These aren't just questions about finding information. They are questions about getting the record right.
For an agent that is only generating text, a relevant passage may be enough.
For an agent that is making a decision, preparing an audit response, investigating an issue or taking an action, the wrong record can lead to the wrong outcome.
Trustworthy answers need to consider identity, relationships, time and source.
That matters because agents are moving beyond generating text. They're being used to investigate, make decisions and take actions.
The usual options still leave the agent with the hard part:
In each case, the agent still has to reconstruct the record.
A Knowledge Pack is a versioned, immutable snapshot that an agent can query directly. There is no query language to invent, no embedding service and no live connection to production. Nothing leaves your environment.
The difference
Agents shouldn't have to rebuild identity, relationships, time and source before they can answer. In a Knowledge Pack, those facts live in the record.
Open the pack. Read its map. Ask for the record.
Who owned this subsidiary on 15 March 2024?
The agent has to reconstruct the answer from fragments.
Six reconstruction steps. Each is another opportunity to get the entity, relationship, date or source wrong.
The record is already structured.
Why packs
System access lets the agent reach source data. It still has to work out what the data means, how it connects, what was true when, and where it was sourced.
Gmail or Drive give an agent access to documents and messages. They don't turn that information into a bounded, queryable record of facts.
Add a tool for every endpoint, and choosing the right tool is where agents slip.
Returns relevant passages, with no guarantee that nothing is missing.
Have full query power, but the schema doesn't explain itself.
Every pack uses the same small set of tools, so adding new knowledge doesn't add new tools. The pack describes itself: what it holds, how records link, and starter queries that show how and where to find records. Records come back structured, with time and source on every link. If anything was cut, the answer says so.
Packs complement Gmail and Drive; they don't replace them.
| Attribute | Data access | Knowledge Pack |
|---|---|---|
| What comes back | Passages or rows | Complete records |
| Structure | Undocumented schema | Self-describing |
| Time and source | Assembled afterwards | On every link |
| Context | Reconstructed by the agent | Built into the record |
| Completeness | Unknown | Stated |
| Tools | One per endpoint | Same small set |
How it works
Open the pack → Read the map → Ask for the record
A standard MCP connector provides agent access across all the major AI platforms.
A local, read-only file or folder. It is not a live database connection.
One read tells the agent what the pack holds, how records link and how to ask. The agent needs no prior training on the dataset.
Describe the record you need and the pack returns it. Follow-up questions stay within the same record.
No query language to write. No search results to piece together.
Properties
Everything the agent needs to know is stored in the record, so it never has to reconstruct identity, relationships, time or source. When an agent is going to make decisions or take actions, getting the record right matters more than finding relevant text.
The agent gets the entity it asked for, rather than having to choose between similar results.
Historical questions can be answered against what was true at the time.
Evidence travels with the record rather than being reconstructed afterwards.
The agent receives the relevant records, not an unrestricted dump of source data.
The agent can distinguish incomplete results from complete ones.
The pack provides hints about where to look next.
Achievements
We've built Knowledge Packs across industrial assets, legal records, corporate entities, geography, venture funding and pharma — from thousands to tens of millions of facts.
The point isn't the size of any individual pack. It's that the same structure and agent interface works across fundamentally different domains. Now we need to prove it on your data.
Demo packs show the model works across domains. They don't show it works on messy production data in your environment. That's what the four-week evaluation is for.
Request a demo See the plant demonstration on Amorfs Read the FAQ
Agents that make decisions and take actions need knowledge they can rely on. Finding fragments of text and guessing what they mean isn't good enough for business. So I built Amorfs.
About us
Pete has spent more than 30 years working at the intersection of enterprise technology, architecture, data and complex business systems, including senior technology roles, at EY, Capgemini, Accenture, UBS and Barclays.
At EY, he spent a decade as Asia Pacific Technology Director for Enterprise Architecture.
He built Amorfs from a simple observation: enterprise systems contain enormous amounts of structured knowledge, but today's AI systems are largely designed to retrieve text rather than reliably reason from the records themselves.
First Cognition Pty Ltd is a privately funded company. We build Amorfs, knowledge infrastructure for AI agents. Amorfs Knowledge Packs put enterprise data directly into an AI agent's hands.
Evaluation
Prove the difference on your own data. A four-week evaluation against 20 real questions, measured against your current approach. This is a paid evaluation.
The result shows whether Amorfs makes a measurable difference on your real questions, with the evidence and context required to trust the answer.
You decide whether the difference is worth taking further.
You agree the questions and answers before anything is scored, so the result doesn't rest on our word.
Amorfs is a fit when an agent needs to work with connected enterprise records and the consequences of getting the facts wrong matter.
If you lead a plant or asset data team, MDM or data governance, compliance or financial crime, legal operations, or another agent programme where getting connected records right matters, start with a twenty-minute demo on the synthetic unit.
Do not send extracts to this website. If we proceed, the extract is versioned in your environment under a signed evaluation agreement.
Run Amorfs against your current approach on a masked extract that never leaves your environment. No production credentials, no cloud ingest, no new platform.
20-minute demo → four-week evaluation → report scorecard
