DragonLairAbout DragonLair
DragonLair is the AI inference control plane between applications, agents, and model providers — one OpenAI-compatible gateway for governed access, evidence-driven optimization, spend, and observability.
Category
Not a silent single-provider proxy and not a pure model aggregator. DragonLair is a control plane: hard governance before ranking, explainable routing decisions, tenant BYOK or prepaid wallet, and a customer portal for keys, usage, and policy.
- Compatible chat, models, and embeddings endpoints.
- Multi-provider routing under cost, capability, and reliability constraints.
- Clear capability boundaries and explainable operational decisions.
Optimizer
Observe → Experiment → Evaluate → Promote or roll back among reviewed choices. Signals include cost, reliability, latency, tool validity, and application outcomes when available.
Governance
Budgets, virtual-model policies, fail-closed guardrails, and explainable decisions via X-Darouter-*. Governance defines the action space before the optimizer ranks.
Browser harness / agent portability
Preview: connect browser and agent runtimes without moving model strategy out of the control plane.