Five Layers
5Top
Retrieval Layer
The answer interface. Validates the requester's content tier, applies a pre-filter to the vector store so only permitted documents enter the candidate set, runs semantic search within that set and returns a cited answer. Two enforcement points: pre-filter before search, post-filter after. Restricted content never enters an uncleared query path.
4Governance
Governance Layer
The honesty engine. Maintains the policy library, the Governance Council model and the Skills Register. Runs human checkpoints at every gate: capability intake, security review, sandbox sign-off and Council approval. Collects the audit log. Monitors content expiry. Never makes a governance decision autonomously.
3Agents
Agents Layer
Specialised roles deployed for specific tasks: CEO review, engineering review, quality assurance, security audit and others. Each agent has a defined role, the tools it may use, the content tier at which it operates and a recommended model. Agents are defined as structured files in the brain registry and deployed on demand.
2Skills
Skills Layer
Approved, versioned automations that the AI can execute. Every skill passes through quarantine, an 8-point security scan, sandbox testing and Governance Council sign-off before it enters the library. Each skill has a trigger description, an owner, a risk rating, a version and a tamper-evident hash. Skills are grouped into licensed packs and distributed to child brains accordingly.
1Foundation
Knowledge Layer
The source of truth. Structured knowledge files with tier tags (1 to 5), classification labels, review dates and expiry fields in every file's frontmatter. Areas, decisions, people, projects and journals. Content is grounded, cited and reviewed on a risk-based cadence. The closed-world assumption applies: the system does not fill knowledge gaps with general training data.
Skill Pack Distribution Model
Source
Master Brain
Canonical skill library. All skills reviewed, approved and versioned here before distribution. The authoritative source of the pack registry.
Base Pack
Core skills every brain needs. Strategy review, decision framework, health check, capability intake and brain audit.
Included at deployment
Delivery Pack
Skills for client engagement and stakeholder management. Stakeholder brief, proposal review and engagement preparation.
Licensed add-on
Customer Growth Pack
Skills for pipeline and lead management. Designed for teams managing a structured sales and nurture process.
Licensed add-on
The distribution rule: Each child brain holds a licensed packs file that lists the packs it has licensed. When skills sync from Master Brain, only skills belonging to licensed packs are delivered. Skills outside licensed packs are never present in a child brain's repository and are never visible to that brain's users. A child brain cannot access a skill it has not licensed, regardless of what exists in Master Brain.
See How Governance Works
Human checkpoints at every gate. Policy library. Closed-world grounding. The full governance model explained.
Trust and Governance