What is semantic dispatch cache?
A semantic dispatch cache is a vector index over compiled skills that routes an incoming goal to an existing skill by meaning rather than by exact string. When an agent asks for something it has effectively done before — even worded differently — the request embeds, matches a cached skill above a similarity threshold, and dispatches straight to deterministic replay instead of re-planning with a language model.
Why it matters
This is the core of Twin’s inverted cost curve. Exact-match or selector-keyed caches only help when the identical action repeats; a semantic cache catches the long tail of re-phrasings, so the share of runs that avoid an LLM call climbs as your agents run more. The result is a marginal cost per run that trends toward zero rather than staying flat.
See it in context: read how Twin compiles and replays a run, follow the cost-cutting guide, or browse use cases and comparisons.