Not a chatbot group. Not a collection of independent agents. A governed digital engineering organization — where expert AI Heads with identity, personality and authority THINK TOGETHER, CHALLENGE TOGETHER, DECIDE TOGETHER with your engineers — and record everything.
The Council is a core capability of the AEGIS Engineering Control Plane. It brings AEGIS core Headings, human engineers, developers, architects and authorized managers into the same engineering session — providing domain positions, challenging one another, identifying conflicts, producing an auditable engineering decision and executing it under AEGIS governance.
This is not a chatbot group, not a collection of independent agents, and not a simple multi-agent orchestration layer. It is a decision-aware engineering system. The right experts, at the right time, producing the right decision — with evidence.
Each Head has a distinct identity, personality, authority domain and voice. Every Head-to-Head message is authenticated and attributable.
Governance Authority — discovers, convenes, moderates, challenges, synthesizes, governs, decides and records. Uses specialist expertise rather than pretending to be the sole expert.
Frontend Engineering Authority — pragmatic, UX-oriented, implementation-focused. Authority over frontend implementation decisions.
Identity Authority — policy-driven access control. Governs identities, authentication, authorization, role boundaries and inter-Head trust.
Security Authority — skeptical, adversarial, risk-oriented. Challenges assumptions and rejects insecure designs, backed by a security veto.
Edge & Deployment Authority — joins whenever edge deployment, distributed execution, network boundaries or offline behavior are affected.
Constitutional & Architectural Authority — AEGIS's architectural memory. Protects AEGIS laws, invariants, boundaries and long-term maintainability.
PHOENIX holds a Resilience & Recovery veto and joins when recovery architecture is impacted. Any AEGIS core heading — VOKA, CRM, ERP, PAYMENTS, WEBSITE and more — can join the council for its domain. VOKA provides voice representation only and never decides alone.
Every council session follows a persisted, explicit state machine — 21 states from CREATED to CLOSED. Discussion is structured, not a free-for-all. Each Head first establishes an independent position with rationale, assumptions, risks, dependencies, recommendation, confidence and evidence — then is cross-examined.
Twenty-plus fields: decision_id, session, mission, participants, positions, objections, risk assessment, selected and rejected options, authority basis, policy basis, dissent, approval requirements and status. A decision must be explainable long after the session.
Code → task → execution plan → decision → council session → discussion → Head positions → evidence — and the reverse. AEGIS answers "why does this implementation exist?" with the relevant Decision Record.
PROPOSED / EXECUTED / VERIFIED / APPROVED are never mixed. No Head may claim completion without evidence. Every task references its session, decision, Head and required evidence.
If the security authority fails, security-sensitive execution is blocked or escalated; if AEGIS fails, execution fails closed. No unrestricted shell or tool capability is granted to a Head by default.
Head authentication, message attribution, scoped credentials, policy enforcement and session isolation are baseline. Tenant A council context never leaks into tenant B.
Guards align with NIST AI RMF and OWASP agent security controls — agent identity, message integrity, least privilege, human approval and circuit breakers.