Architecture & Infrastructure6 min read
Memory Without an LLM in the Loop
Eliminating external language models from the memory write path using local zero-shot ONNX models, FastEmbed, and deterministic civil date resolution to build sovereign, cost-predictable agent memory.