Synapse-Control Reduces Agent Inference Costs with 1.7B Classifier
November 11, 2026
Synapse-Control uses a 1.7B classifier to select advice for LLM agents instead of generating new diagnoses. This lightweight control plane uses micro-contracts and a Prompt Value Table to achieve 29.8% success in AppWorld and WebShop benchmarks.
HOW THIS AFFECTS YOU
●
builderYou can lower the cost of agent experience reuse by using a small classifier instead of calling large models for every failure diagnosis.