Hi HN. I'm from Llewellyn Systems. ODE (Organized Decision Engine) is an AI-native enterprise OS — not a chatbot wrapper, not an agent framework. It's a runtime

Hi HN. I'm from Llewellyn Systems. ODE (Organized Decision Engine) is an AI-native enterprise OS — not a chatbot wrapper, not an agent framework. It's a runtime where 21 autonomous AI agents coordinate under enforceable governance. It's live in production. Why an OS, not a point solution: Most AI-in-enterprise approaches bolt a model onto one workflow. We built a shared execution environment where agents have defined skills, scoped permissions, and structured communication. 21 agents, 55 registered skills, 800+ API endpoints. Agents operate within a skill chainer that composes multi-step executions across agent boundaries. Constitutional governance — hardcoded, not prompted: ODE enforces 10 constitutional laws. These aren't system prompts. They're hardcoded constraints checked on every agent action at the application layer. No agent can authorize its own financial exposure. No single agent can override multi-tenant data isolation. Every decision must produce a verifiable audit record. 3-head decision protocol: High-risk actions require sign-off from three executive agents evaluating orthogonal dimensions: operational impact, constitutional compliance, and financial exposure. All three must approve. Inspired less by AI safety literature and more by nuclear launch protocols — independent verification across different concern axes. Crypto-backed provenance: Every agent decision produces a cryptographically verifiable audit trail. In regulated industries, "the AI decided" isn't acceptable. Agent-to-agent commerce: External AI agents can discover ODE's capabilities via .well-known/agents.json, then hire and pay ODE agents through an A2A protocol. Not agent orchestration within a system — agent commerce across organizational boundaries. What's open source: 15,000 enterprise use cases on HuggingFace: https://huggingface.co/datasets/LlewellynSystems/ode-enterprise-use-cases https://huggingface.co/datasets/LlewellynSystems/ode-enterpr... Stack: Next.js 15, Express.js 5, Supabase PostgreSQL, Prisma ORM. Honest bottleneck: Latency on 3-head consensus under load — p99 gets uncomfortable on complex actions. Agent commerce dispute resolution between organizations is a hard open problem. Happy to go deep on any of this.