Skip to content
DATTA
Information comes in
Concept visualization

DATTA / Technology

Inside a brain of your own.

The architecture behind memory, permissions and verifiable operations.

The architecture, in detail.

01 / Technology

Four levels of knowing.Almost the whole market sells the first two.

An AI can search over a graph. To act, it needs the fourth level.

DATTA
Knowledge takes shape
Concept visualization

02 / Information comes in

Memory stays. The AI can change.

Documents, decisions and their context live in the vault. The model receives authorized context for its task; it does not own the memory.

This separation preserves your history when you switch assistants. A company should not have to rebuild its knowledge in every conversation.

POLICYv.04
v.01v.02v.03v.04
Source and authorVersion and validitySupporting evidence

03 / Knowledge takes shape

One fact. Its entire history.

Knowing what was said is not enough. The source, who stated it, when it became valid and its supporting evidence matter.

A new version preserves its predecessors. When a decision lacks a source, the brain recognizes the gap and asks for it instead of treating an assumption as certainty.

MCP
  1. 01Request
  2. 02Check permission
  3. 03Return authorized context

04 / Permission sets the boundary

Context travels with permission.

People and digital employees receive the context their role needs. Compatible assistants access it through MCP, with grants that limit categories and time.

Connecting a tool does not open the entire vault. Revoking a grant stops new access. Sensitive information stays closed unless explicitly authorized.

Security
DATTA / Concept visualization
Source and authormeeting-04
Version and validityv.04
Supporting evidencepolicy-03
Scope and durationread · 30 min
Operation recordsha256 · 8f21…c7a0
Everything leaves a receipt

05 / An operation leaves evidence

Every operation ends in evidence.

A read, write or decision records who requested it, what was authorized and when it happened.

Entries are chained so changes to the past can be detected. Digital employees prepare work; a person confirms execution. The result returns to memory.

DATTA / SPECS

The architecture, in detail.

You are being asked to trust a place with the most sensitive information of your life. So here is the whole picture, including the parts that are not finished.

Encryption
Your vault is encrypted in transit and at rest. Keys are held under a documented custody procedure, separate from the data.
Live today
Where your data lives
On servers we operate directly, today in the United States, not on a managed cloud service that holds the keys for us. The stack is portable by design, so where it runs can change without the product changing.
Live today
Permissions
Nothing is shared by default. Every grant names what it covers, has an end date, and can be revoked at any moment.
Live today
What an AI can reach
An assistant reads only the categories you granted, for as long as you granted them. The most sensitive levels stay closed unless you open them.
Live today
Per turn consent for the most sensitive
Opening a sensitive item for one answer only, and closing it again when the answer is done.
In development
Audit log
Every read and every write is recorded with who asked, what was granted and when. Entries are chained, so a change to the past does not go unnoticed.
Live today
Never training data
Your vault is not used to train models, not by us and not by the providers we work with.
Live today
Export and deletion
Exporting everything and deleting your account are part of the account, not a support ticket.
Live today
Identity verification
Against the document itself, including the chip in passports and modern ID cards, so what is verified is verified against the source.
Live today
Independent audit
Not done yet. When an external audit happens, the result is published here, whatever it says.
In development

DATTA / Technology

Talk with Datta

We’re exploring: Technology.