Project Liberty AIOS
How it works, one page
Concept · placeholder brand "Beacon"
An AI operating system for Liberty advisers, built on Microsoft, running in Liberty's own tenant. Advisers get a plain-English assistant over their whole book, and every action is drafted for their approval.
How we build it (once)
VS Code + GitHub Copilot / Codex
Our workshop, where the system is authored. Advisers never see it and it is not part of what runs in production.
▼
The operating system, the IP
Context · data · skills · instructions (file-native)Proven on a GPT model, 10 Jul
The durable value. Model-agnostic by design, which is why it runs unchanged on Microsoft's stack. Verified: our exact architecture ran under GitHub Copilot on a GPT coding model, reading context, running automations and following workflows first time.
▼
What advisers use (always-on, in your tenant)
Azure AI Foundry agent + Azure OpenAI, delivered in Microsoft Teams / a Power Apps portal
The hosted agent reads the operating-system files and answers in Teams or a branded portal. This is the dashboard's backend: Liberty's cloud, always on.
How it connects to an adviser's world
Native (Microsoft Graph)
Because it lives inside Liberty's Microsoft tenant, these connect out of the box, per adviser, each seeing only their own data.
Outlook · Teams · Calendar · SharePoint
API connector
Standard connectors into the systems that expose an interface, read and write, governed centrally.
CRM / Adviser Workbench
File ingestion
Blueprint has no API. The export an adviser already downloads is dropped into a watched folder and read into the assistant's knowledge. File-first is exactly what the architecture is built for.
Blueprint exports (SRS, policy)
Built for a tier-one financial services environment
In-countryAzure OpenAI in South Africa North, inside your tenant.
POPIASection 72 covered via Microsoft's data agreement.
Per-adviser isolationEntra identity, each adviser sees only their book.
Human in the loopNothing sends without the adviser's approval.
What Liberty's own build takes quarters to stand up, we can stand up in days, on your stack, your models, in your tenant.