The Manifesto

AI agents are dumb like sheep. Out of the box, they wander, hallucinate, forget, contradict themselves between sessions. No memory. No boundaries. No chain of command.

The industry's answer? Make the sheep bigger. More parameters. More context. More tokens.

Size does not solve governance. A larger sheep is still a sheep.

The answer is not a bigger sheep. The answer is a shepherd.

I. Statute is Law

Agents don't improvise. They execute written rules – or they escalate.

The Statute is the authoritative description of how work is done. Not a suggestion. Not a guideline. Law. Tickets, chats, and emails can propose changes; they cannot override Statute.

The rules evolve – every serious failure either updates the Statute, adds a case, or both. But they are always written down.

II. The Human Commands

One human owns every decision. Agents propose. Agents execute. Agents review. They never decide.

Agents operate inside narrow Rules of Engagement. They do not redefine goals, values, or strategy. That is the shepherd's job.

The human is accountable for every material action performed by agents. Not the model. Not the framework. The human.

III. Own Your Stack

Your data stays yours. Knowledge Lake lives in a git repo you control – not in someone else's SaaS. Observability runs on open-source – OpenTelemetry, Prometheus, Grafana – not on a Datadog invoice. Local models where it makes sense.

Banks won't store secrets in Notion. Pharma won't pipe patient data through a third-party API. Neither should you.

What goes to the cloud stops being yours. Your secrets stop being secrets. This system is built for those who understand that – and for those who can't afford to learn it the hard way.

Open-source. Self-hosted. Sovereign.

IV. The Eye Sees Everything

Every agent action is logged. Every tool is a named gate with a risk level and a paper trail. Read-only by default – mutations require explicit Statute and human approval. Every serious failure grows the Statute.

You cannot command what you cannot see. And this system sees everything.

V. Small and Sharp

Capability is modular. Each skill is scoped, testable, and replaceable. Complex workflows are compositions of small skills – not monolithic mega-agents.

A squad of specialists beats one generalist every time. Commander Shepard knew this.