The Council Protocol: 81 Rules for Democratic AI Governance

Autonomous AI agents making real decisions is no longer hypothetical. Agents write code, deploy infrastructure, and modify databases. The question is no longer "can they act autonomously?" but "who governs what they do?"

Most frameworks answer with guardrails — hardcoded rules that block specific actions. Do not delete production data. Do not deploy without tests.

Guardrails are necessary. They are not sufficient. They tell agents what they cannot do. They say nothing about how agents should decide among the things they can do.

Codmir's Council Protocol fills this gap. Not a list of prohibitions — a constitutional framework of 81 rules derived from first principles, governing how decisions are proposed, debated, justified, and recorded.

The Problem with "Trust Me"

When a single AI agent makes a decision, you get an output. Maybe it explains its reasoning. Maybe not. A black box produced an answer. Accept or reject.

This works for low-stakes suggestions. It does not work when agents ship code to production or choose architectural patterns that constrain future development.

You need to know: What was the decision? Who voted? What rules applied? What alternatives were considered?

Corporate governance solved this centuries ago. Board votes are recorded. Bylaws constrain decisions. Audit trails preserve reasoning. Codmir agents operate under the same standard — at machine speed.

The 81 Rules: Seeds, Aspects, and Contexts

The Council Protocol derives its rules mathematically from three dimensions.

Five Seeds

The Seeds are foundational principles — constitutional bedrock.

Covenant governs agreements and commitments. When agents commit to a task, what obligations does that create?

Chronicle governs memory and record-keeping. What must be recorded? How long must records persist? This ensures decisions remain auditable long after they are made.

Boundary governs scope and limits. Where do agent authorities end? Not as hardcoded walls, but as principles that adapt to context.

Guardian governs safety and protection. When actions carry risk, what safeguards apply? Autonomy never comes at the cost of safety.

Network governs communication and coordination. When is consensus required versus unilateral action? This defines the social contract between agents in the mesh.

Nine Aspects

Each Seed expresses through nine Aspects — dimensions like authority, transparency, precedent, proportionality, reversibility, consent, efficiency, equity, and accountability.

The Guardian seed through the reversibility aspect asks: "Can this action be undone? If not, what additional governance applies?"

Nine Contexts

Each Seed-Aspect combination adapts across nine operational contexts: sprint execution, code review, deployment, resource allocation, conflict resolution, external communication, data handling, security operations, and emergency response.

5 Seeds x 9 Aspects x 9 Contexts = 405 intersections, distilled into 81 governing rules. This is not arbitrary complexity. It is structured completeness — the 81 rules cover the governance space the way a periodic table covers elements, with each position defined by its coordinates.

How Decisions Work

When a decision point arises, the Council Protocol activates.

Representatives Assemble

Representatives are agents or humans with voting authority for the decision at hand. The protocol determines relevance based on scope and invoked rules. A database schema decision invokes different representatives than a deployment decision. Governance is contextual.

Voting with Confidence Scores

Representatives do not simply vote yes or no. Each vote carries a confidence score reflecting certainty. A domain expert votes with high confidence. A tangentially involved agent votes lower. The weighted outcome reflects informed majority opinion.

Rule Citations

Every vote must cite specific rules: "I vote to escalate this deployment because Rule 37 (Guardian / Reversibility / Deployment) requires human approval for irreversible production changes." Guardian-seed rules apply higher weight to safety specialists. Network-seed rules weight coordination experts more heavily.

This is not bureaucracy. It is accountability.

The Audit Trail

The outcome — decision, vote breakdown, confidence scores, rule citations, dissenting positions — is recorded permanently.

Six months later, when someone asks "why did the agents choose this architecture?" the answer is: "Representatives A, B, and C voted for approach X with confidence scores of 0.9, 0.7, and 0.85, citing Rules 12, 44, and 67. Representative D dissented, citing Rule 23. The weighted outcome favored X by 3.2 to 1.1."

That is governance. That is what makes autonomous AI trustworthy.

A Living Protocol

The Council Protocol is designed to evolve. New rules are proposed through the protocol itself — representatives vote on amendments using the same governance framework. Edge cases reveal gaps and generate amendments. Overly restrictive rules get refined. The governance system governs its own evolution, at machine speed.

See Governance in Action

Autonomous AI without governance is a liability. Autonomous AI with transparent, auditable, constitutional governance is an asset. The Council Protocol makes every agent decision traceable, justified, and constrained by principle — not by brittle hardcoded rules, but by a living framework that evolves with use. Explore Codmir and see what it means to trust AI decisions not because you hope they are right, but because you can verify exactly why they were made.