
very catastrophic hedge fund failure — LTCM, Archegos, FTX — shares one root cause: the risk function was structurally powerless, overruled by the people making money. AEGIS Fund inverts this. It's a fully autonomous hedge fund run by 11 AI agents — a CIO, competing portfolio managers, traders, sealed strategy bots, and independent CRO, Compliance, and Research officers — each on a different framework (LangGraph, NVIDIA NIM / Llama 3.3 70B, and deterministic adapters), coordinating entirely through Band. Band is the collaboration layer, not a wrapper. Every directive, report, advisory, block, and halt is a Band message, routed by @mention through shared rooms and permanently logged as a unified audit trail. No agent calls another directly; remove Band and the system cannot function. The command chain flows top-down (capital mandates) and bottom-up (aggregated reports), while the CRO and Compliance agents cut across it with hard-coded authority no portfolio manager or CIO can bypass. The strategies stay sealed — the governance layer supervises bots it cannot see, with only abstract risk signals crossing the boundary. In the live dashboard, the fund runs eight market ticks: at tick 5 Compliance blocks a desk on a rules violation; at tick 7 the CRO halts the entire fund on a drawdown breach — two independent overseers, different mandates, with zero human intervention. AEGIS is more than a finance demo. It's a blueprint for any high-stakes workflow — lending, claims, compliance review — where oversight exists on paper but bends under pressure. It shows that multi-agent systems can have enforceable, auditable authority structures, coordinated across frameworks through Band.
19 Jun 2026