Eleven-agent council
Socrates hub + Veridian, Ceres, Gleaner, Argus farm + Frog, Lisa, Eagle, Ezra, White Rabbit specialists + Glaucon, the loyal opposition. Selective activation -- no agent forced to speak every turn.
MetaGen Auto
Autonomous Multi-Agent
AI Operations
An eleven-agent council orchestrates research, build, validation, and self-improvement under a human-in-the-loop merge boundary. Built on Next.js 16 + AG2 0.13.4 + a star hub-and-spoke topology.
Star hub-and-spoke
Socrates dispatches via @register_for_llm tools. Spokes return briefs against a Pydantic schema. Sub-patterns from the AG2 cookbook compose without forking.
HITL merge boundary
AI agents author freely; humans approve at seven locked gates -- phase transitions, mutation acceptance, production deploys, tool installs, policy edits, A2A exposure, story promotion.
Self-improvement engine
P7 mutation cycle proposes diffs against the codebase. Every mutation runs the full eval slate before landing. Mutations are never auto-merged -- the CHANGELOG flips on human review.
AIDD + TDD discipline
AI agents author the inner loop. Every artifact lands behind a failing eval/test first and turns green as the implementation is built. Five-places-one-phase enforced.
Gate-locked verification
P0-P11 complete; P12 active. Every change lands behind a failing eval case first -- structural, per-phase, web-story, and live gates run on each push, all green; mutation_cycle rejects regressions automatically.
Security
Public landing — operations console behind sign-in
Press Play, ticket CRUD, HITL approval gates, mutation review, the multi-agent council chat, mission control, artifacts, logs, and settings are restricted to authenticated operators. No operational state is rendered on this page.
NEXT_PUBLIC_AUTH_REQUIRED + NEXT_PUBLIC_AUTH_DISABLED env flags control the gate