🛡️Sovereign Mode (Zero Egress)Complete privacy by design. All intelligence nodes, raw documents, and trained weights remain strictly on your Mac's filesystem.
⚡Apple MLX AccelerationRuns neural weights directly inside VRAM using Apple's official MLX framework, leveraging standard unified memory for sub-10ms response times.
🧠Specialized Micro-BrainsPluggable micro-models (<1M parameters) running persistently as background daemons to execute search, classification, and style synthesis.
🗡️Intent Blade OrchestratorA quantized local SLM acting as a Grounded Agent, translating natural language into SQL filters, entity aggregations, or semantic queries.
🔨Teacher-Student ForgeIn-house distillation pipeline using local Mistral-7B to automatically label, format, and train light-weight student micro-models.
💾Isolated Node StorageUnique SQLite databases and .npz indices per intelligence node to guarantee zero lock contention and infinite horizontal scale.