A runtime for AI agents: define tools, route requests across models, and stream output back to the client — built to run many tenants on one engine.

Agent Orchestrator is the engine behind Lumi and Moliam's product surface — a multi-tenant runtime that turns model calls + tools into reliable, observable agents.

Each tenant defines its own tools and policies; the router picks a model per request, retries and falls back on failure, and streams tokens to the client over a resilient socket layer. Runs and tool calls are traced end-to-end so behavior is debuggable in production.