Docs
Enzyme compiles content into a concept graph — under 15 seconds for 1,000+ documents. Queries run on-device in ~8ms. No conversation history needed. These pages explain how the engine extracts cross-cutting connections, what the agent receives, and why it works whether the content is a personal vault or a product's user corpus.
Setup
Install, point at your notes, explore.
How it works
The compile step: from content corpus to searchable concept graph.
Arriving knowing the room
The LSP analogy: what it means when your agent has structural understanding before the conversation starts.
Catalysts
Pre-computed questions that cut across your content — the search layer that keyword and vector search can't reach.
Configuration
The config.toml reference — entity selection, embedding limits, and provider setup.
Apply
Project your concept graph onto an unfamiliar corpus. Your lens, their content.
How memory compounds
Temporal structure, questions over conclusions, and what happens when agents write back.
In practice
Short clips from real work sessions where enzyme changed what the agent found.
SDK Preview
Private betaYour users accumulate taste through annotated recipes, curated collections, logged experiments. Enzyme turns that accumulation into an agent that helps them go deeper — try things they wouldn't have thought to try, see patterns in their own practice, and build on what they've already learned.
SDK preview
What happens when an agent can read the accumulation — and ask questions back.