Self
Intelligence

Internal draft London October 2026 Self

Contents

  1. 01What it is
  2. 02How it works
  3. 03Benefits

01 What it is

A system of intelligence.
Automate. Build. Learn.

A knowledge layer under the company’s stack. One memory that every agent shares: kept current, within access rights, giving each task only the context it needs, with sources.

The gap

AI that knows the company needs the company’s data. That creates two problems.

  • A Security risk Company data leaves the perimeter. Prompts, documents and answers pass through public models run by third parties, outside the company’s control: exposure to retention, leakage and compliance breaches.
  • B Cost Spend grows with every user and every agent. The same context is sent, and paid for, on every call, and what the company already knows is rediscovered on every question.

What it is built from

PEOPLE AND APPS API · MCP · SQL External models INSIDE THE COMPANY’S PERIMETER Gateway checks who may see what Agent orchestrator gives each task only the context it needs AGENTDevelopment AGENTSales AGENTSupport AGENTAnalytics AGENTCompliance ASK CONTEXT, WITH SOURCES Corporate memory one graph of facts, decisions and people every fact keeps its source and when it was true what one agent learns, every agent remembers UNDERSTOOD ONCE, WHEN WRITTEN Data layer stores, indexes, keeps current EVENTS AS THEY ARRIVE Connectors one adapter per source vLLM gateway one door to the models ONLY IF ALLOWED Local models PRINCIPLES Kept current Within access rights Only the context needed Answers with sources MAIL CHAT MEETINGS DOCS CRM TICKETS WHERE THE WORK HAPPENS

The solution

This architecture solves both.

  • A Data stays inside Everything runs inside the company’s perimeter. A gateway checks who may see what, local models do the work, and external models are used only if the company allows it.
  • B Context paid for once Meaning is fixed once, when a fact is written, and one memory serves every agent. Each task gets only the context it needs, and routine work runs on local models.

02 How it works

One memory under every agent.

Self Intelligence reads what the company knows as it happens, understands it once, when it is written, and keeps it in one graph that people and agents share. What one agent learns, every agent remembers.