A constellation of domain-expert AI models — vision, audio, video, timeseries, text — orchestrated through an intelligent inference router. No monolith. No retraining. Just fast identification, continuous learning, and structured context.
Classify incoming data. Route to the right expert model. Collect structured context. Enrich from memory. Escalate to an LLM when reasoning is needed. Store everything. The system gets smarter with every request.
Every database type the platform needs. Dual purpose: persistent storage for device and sensor data, and memory for Earth's models — embeddings, knowledge graphs, historical patterns, state.