Preemptive Context Layer (PCL): from per-tool retrieval to a unified hybrid index for LLM-based personal assistants

Personal assistants built on language models usually reach a person's data through tools: the model reads a message, decides which services might hold something relevant, and queries them. Whatever the model does not think to look for is never seen. This note defines the Preemptive Context Layer (PCL), in which everything the assistant is connected to is kept in one continuously updated index, a single live picture of the person's world, and that index is searched automatically, by meaning and by keyword at once and without any language-model call, before the model sees each message. The model begins with the relevant context instead of spending tokens and tool calls deciding where to look, reads detail from the assistant's local copy of the data, and calls a live service only when that copy is not enough. The note describes the reference implementation in ClosedHand, an open-source personal assistant, states its limits, and places it relative to retrieval-augmented generation and enterprise grounding. The name is free for anyone to use. First published at https://closedhand.com/pcl. Version 1.1 adds depth on demand (section 3.6).