We build the infrastructure layer that keeps the agentic internet human.

A web of data that people and agents can find, use and own.

Truth is at risk.
It is being compressed into model weights.

A model learns from the world's knowledge and stores it as billions of numbers. No person can read those numbers. No one can check a fact inside them, correct it or name its source.

We keep knowledge outside the weights, as facts and rules with their sources. People can read them. Any model can query them.

The semantic web

The semantic web needed machines that read and write.
They exist now.

The idea is 25 years old. It stalled because people had to write every fact. Language models can write them now.

1989 and 2001

From a web of pages to a web of data.

Tim Berners-Lee built the web for people to read. In 2001 he proposed a second web: facts that machines can read.

2004 to 2017

The W3C wrote the standards.

RDF records a fact. OWL says what it means. SPARQL asks questions. SHACL checks the data.

Since then

People had to write the facts by hand.

That was slow and costly. So most knowledge stayed in documents that only people can read.

Now

Models write the facts. Agents use them.

A model reads a document and writes its facts. An agent queries them and cites the source.

Where this goes

Five things will be true.

All data will be interoperable and perfectly recalled.

Step 1 of 4

1Five systems describe Carol. Each one uses its own shape.

CRMContact

FirstNameCarol

LastNameIbrahim

Emailcarol@acme.com

HREmployee

emp_idE-4471

full_nameC. Ibrahim

managerE-0912

ERPApprover

appr_id88

approvedJE-1042

ChatUser

handle@carol

displayCarol I.

Webfoaf:Person

foaf:nameCarol Ibrahim

Five shapes of one person

2Links say which records are the same person, and which fields mean the same.

crm:0031sameAshr:E-4471sameAserp:88

CRMFirstName + LastName→:name

HRfull_name→:name

Webfoaf:name→:name

CRMEmail→:email

HRmanager→:reportsTo

ERPapproved→:approved

3One person. Each field keeps its source.

One person

:CarolIbrahim  a  :Person ;

    :name       "Carol Ibrahim" ;

CRM · HR · WEB

    :email      "carol@acme.com" ;

CRM

    :reportsTo  :BobKlein ;

    :approved   :JE-1042 .

ERP

Five records, one person

4One question reads across all five systems.

Who approved JE-1042, and who is their manager?

JE-1042approvedByERPCarol IbrahimreportsToHRBob Klein

Answered from ERP and HR in one query

Law and regulation will be codified.

Step 1 of 4

1A law, as it is written.

Article 33 · Notification of a personal data breach to the supervisory authority

1. In the case of a personal data breach, the controller shall without undue delay and, where feasible, not later than 72 hours after having become aware of it, notify the personal data breach to the supervisory authority competent in accordance with Article 55, unless the personal data breach is unlikely to result in a risk to the rights and freedoms of natural persons. Where the notification to the supervisory authority is not made within 72 hours, it shall be accompanied by reasons for the delay.

2. The processor shall notify the controller without undue delay after becoming aware of a personal data breach.

3. The notification referred to in paragraph 1 shall at least:

(a) describe the nature of the personal data breach including where possible, the categories and approximate number of data subjects concerned and the categories and approximate number of personal data records concerned;

(b) communicate the name and contact details of the data protection officer or other contact point where more information can be obtained;

(c) describe the likely consequences of the personal data breach;

(d) describe the measures taken or proposed to be taken by the controller to address the personal data breach, including, where appropriate, measures to mitigate its possible adverse effects.

4. Where, and in so far as, it is not possible to provide the information at the same time, the information may be provided in phases without undue further delay.

5. The controller shall document any personal data breaches, comprising the facts relating to the personal data breach, its effects and the remedial action taken. That documentation shall enable the supervisory authority to verify compliance with this Article.

Regulation (EU) 2016/679 (GDPR), Article 33. Source: EUR-Lex.

2The same text, broken into parts an agent can use.

Article 33 · Notification of a personal data breach to the supervisory authority

1. In the case of a personal data breach, WHOthe controller shall without undue delay and, where feasible, WHENnot later than 72 hours after having become aware of it, MUSTnotify the personal data breach to the supervisory authority competent in accordance with Article 55, UNLESSunless the personal data breach is unlikely to result in a risk to the rights and freedoms of natural persons. IF LATEWhere the notification to the supervisory authority is not made within 72 hours, it shall be accompanied by reasons for the delay.

2. The processor shall notify the controller without undue delay after becoming aware of a personal data breach.

3. The notification referred to in paragraph 1 shall at least:

(a) WITHdescribe the nature of the personal data breach including where possible, the categories and approximate number of data subjects concerned and the categories and approximate number of personal data records concerned;

(b) WITHcommunicate the name and contact details of the data protection officer or other contact point where more information can be obtained;

(c) WITHdescribe the likely consequences of the personal data breach;

(d) WITHdescribe the measures taken or proposed to be taken by the controller to address the personal data breach, including, where appropriate, measures to mitigate its possible adverse effects.

4. Where, and in so far as, it is not possible to provide the information at the same time, the information may be provided in phases without undue further delay.

5. ALWAYSThe controller shall document any personal data breaches, comprising the facts relating to the personal data breach, its effects and the remedial action taken. That documentation shall enable the supervisory authority to verify compliance with this Article.

Regulation (EU) 2016/679 (GDPR), Article 33. Source: EUR-Lex.

3Each part becomes a rule, stored as data.

Whothe controller33(1)

Mustnotify the supervisory authority of the breach33(1)

Whenwithin 72 hours of becoming aware33(1)

Unlessa risk to people's rights and freedoms is unlikely33(1)

If lategive the reasons for the delay33(1)

Withnature · contact point · consequences · measures33(3)

Alwaysdocument the breach, its effects and the remedy33(5)

4An agent decides a case, and acts.

Casebreach #0412

foundMonday 09:10

records1,240 customers

risk to peoplelikely

Notify the authority: yes, a risk is likely.33(1)

Deadline: Thursday 09:10, 72 hours after Monday 09:10.33(1)

The draft has all four parts.33(3)

The breach is on record with facts, effects and remedy.33(5)

Each decision cites its paragraph

People and organizations will truly own their knowledge and use its full potential.

Step 1 of 4

1Your knowledge is a graph of facts.

2One query takes all of it out.

Your query

CONSTRUCT { ?s ?p ?o }
WHERE     { ?s ?p ?o }

graph.ttl

:JE-1042  :approvedBy  :CarolIbrahim ;
          :preparedBy  :BobKlein ;
          :follows     :Policy-4.2 .
:CarolIbrahim  :reportsTo  :BobKlein .

3Any tool that reads RDF reads the file.

a SPARQL storea notebookanother vendor

Open standards: RDF, Turtle, SPARQL

4The history of each fact stays.

commit 409:JE-1042 :preparedBy :BobKlein

commit 412:JE-1042 :approvedBy :CarolIbrahim

as of commit 409: who approved JE-1042?

Not yet approved, as of commit 409

Public data will be discoverable, usable, embedded and classified.

Step 1 of 4

1Public datasets, side by side.

DBLP566.2M triples

Wikidatageneral facts

UniProtproteins

YAGO 4people and places

2Each fact gets an embedding and a class.

DBLPrecord

titleKnowledge Graphs

venueACM Computing Surveys

year2021

Embedding

Class:Paper

3A search finds facts by meaning.

how machines store what they know

:PaperKnowledge Graphs2021

:PaperIndustry-scale Knowledge Graphs2019

:PaperKnowledge Graph Embedding: A Survey2017

Found by meaning

4One query joins public data with yours.

Your report Q3citesyoursKnowledge GraphspublishedInDBLPACM Computing Surveys

3 of your reports cite these papers

Every model of the world will cover all media, readable by people and checked by machines.

Step 1 of 5

1A PDF, as it is: text, a figure and a table.

recall-notice.pdfpage 1 of 4

Recall notice RN-2018-07

Lot 73091 of valsartan is recalled. An impurity was found above the limit. Three finished batches contain the lot.

Figure 1 · Impurity by lot

limit73088730897309073091

Table 1 · Batches that contain the lot

Batch	Site	Tablets
CVX-A17	Berlin	12,000
CVX-B04	Lyon	8,400
CVX-K02	Vienna	9,600

2The same page, broken into its parts.

recall-notice.pdfpage 1 of 4

TextRecall notice RN-2018-07

LOTLot 73091 of valsartan is STATUSrecalled. An impurity was found WHYabove the limit. HOW MANYThree finished batches contain the lot.

FigureFigure 1 · Impurity by lot

limit73088730897309073091

TableTable 1 · Batches that contain the lot

Batch	Site	Tablets
CVX-A17	Berlin	12,000
CVX-B04	Lyon	8,400
CVX-K02	Vienna	9,600

3The parts join the graph. Each fact keeps its source.

One page, three kinds of content, one graph

4A person can read the graph. Nobody can read model weights.

The same, inside model weights

-0.9999   0.3154   0.5561  -0.4135
 0.3277  -0.8104  -0.5296  -0.2114
-0.2070   0.3469   0.8350   0.1942
-0.6903  -0.6543  -0.4654   0.2970
-0.2885  -0.9230   0.8342  -0.3316
-0.8251  -0.1323   0.8898   0.3044
-0.5383   0.2693  -0.0804  -0.4608
-0.8400   0.0119   0.1032   0.6220
-0.3755  -0.5254   0.3608   0.2823

In the graph

Lot 73091 is recalled.

Its impurity is above the limit.

Three batches contain it.

They are in Berlin, Lyon and Vienna.

5A machine checks the graph across the parts.

The text says three batches. The table has three rows.

The lot in the figure is the lot in the text.

Each batch in the table names a site.

Text, figure and table agree

Who we are

We are little big brain,
a company in Berlin.

We build one product: a search platform for agents and humans. It keeps knowledge as a graph in object storage. The SDKs and the MCP server are open source.

Company details in the imprint

Based in: Berlin, Germany
Founder: Rino Montiel
Contact: hello@littlebigbrain.com
Code: github.com/littlebigbrains
Follow: LinkedIn · X

Join us

Build it with us.

The SDKs and the MCP server are open source on GitHub. Follow the work on LinkedIn and X.

We build the infrastructure layer that keeps the agentic internet human.

Truth is at risk.It is being compressed into model weights.

The semantic web needed machines that read and write.They exist now.

From a web of pages to a web of data.

The W3C wrote the standards.

People had to write the facts by hand.

Models write the facts. Agents use them.

Five things will be true.

All data will be interoperable and perfectly recalled.

Law and regulation will be codified.

People and organizations will truly own their knowledge and use its full potential.

Public data will be discoverable, usable, embedded and classified.

Every model of the world will cover all media, readable by people and checked by machines.

We are little big brain,a company in Berlin.

Build it with us.

Truth is at risk.
It is being compressed into model weights.

The semantic web needed machines that read and write.
They exist now.

We are little big brain,
a company in Berlin.