Cardinal

One ring of agents.
One human at the centre.

Cardinal is the human window onto an engagement that a ring of long-lived AI agents runs day and night. The agents scan wide and metabolise the daily storm of signal into structured artifacts; the human makes the decisions that carry weight. The database between them is the single source of truth — and it remembers everything.

Agents metabolise · humans decide One ring · a few roles · a family of scouts Every write audited
The Cardinal mascot — a geometric red cardinal in flight over an open compass

What is Cardinal?

An engagement generates more signal than any one person can read — hundreds of messages, meetings, backlog changes and documents a day. Cardinal splits that work in two. A ring of AI agents watches everything, groups and summarises it, and drafts the briefs, decisions and replies. A human reads a short, ranked picture and makes the load-bearing calls — approve, ratify, assign, publish. Cardinal is where that human half happens, and the tamper-evident database is the contract that keeps the two halves honest.

The problem it solves

No human can stay on top of an engagement's full signal stream and still do deep work. Triage eats the time that should go to judgement.

The idea

Let agents do the broad, shallow metabolising; reserve the human's attention for the narrow, deep decisions. Make every step traceable so the whole thing can be trusted.

Three ways in

This documentation is organised around three pillars. Start wherever your question lives — or read How it works first for the whole story in one page.

The application

The web app the human uses — its pages, its front end and back end, and the data model that is the system of record.

The agents →
The data model →

The agentic fabric

The ring of agents that drives it all — who they are, how they coordinate, the scouts that feed them, and the cadence that keeps them ticking.

Meet the agents →
See it run →

Install & infrastructure

How it is set up and where everything lives — the desktop host, the shared Azure substrate, and the per-engagement slice.

Install →
Infrastructure →

At a glance

How it behaves

  • Agents are read-wide, write-narrow — each owns only its own tables
  • Agents never message each other directly; they coordinate through the database
  • Nothing reaches the customer without a human pressing Publish
  • Every write — agent or human — is hash-chained and tamper-evident

What it's built on

  • A React web app over a typed API on Azure
  • A Postgres database as the single source of truth
  • A desktop AI host running the long-lived agents
  • A shared Azure substrate with a thin per-engagement slice
New here? The fastest way to understand Cardinal is to follow one signal from arrival to a human decision. That's the How it works page — about a five-minute read.