Apply

In addition to looking inward — your workspace, your entities, your catalysts searching your content — Enzyme can also point the lens outward.

enzyme apply ./research-papers

This indexes an external directory using your workspace’s catalysts. Your questions, pre-computed against their content. Then you search:

enzyme catalyze "fairness in algorithm design" --vault ./research-papers

The search runs through your catalysts, not the papers’ vocabulary. Your workspace’s question — maybe something like “How do constraints reveal possibilities?” — finds papers that resonate with how you already think. Papers you’d have found interesting but wouldn’t have known to search for, because they use different language for the same concerns.

What’s happening underneath

Apply runs the same pipeline as init, but split across two databases. Your workspace keeps its catalysts. The external directory gets indexed, embedded, and similarity-mapped against your existing catalysts. The search at query time reads catalysts from your workspace and similarities from the target corpus.

The external content doesn’t need tags, links, or any structure. It just needs to be text. Markdown, plain text, whatever your directory contains. Enzyme chunks it, embeds it, and measures how each chunk relates to each of your catalysts.

When to use it

Apply works whenever you have accumulated thinking in your workspace and unfamiliar material you want to explore through that lens.

A researcher with years of notes on ethics in technology downloads 50 papers on algorithmic fairness. Apply projects their existing intellectual framework onto the new papers. The search surfaces papers that speak to their concerns, even when the papers use academic language their notes never would.

A team lead with months of meeting notes and decision records applies their workspace to a new hire’s onboarding docs. The catalysts surface where the new material connects to ongoing threads — and where it doesn’t, which is sometimes more interesting.

Apply works on any text corpus. If your domain’s data can be represented as text — interaction logs, save histories, annotated collections, meeting transcripts — your catalysts can search it. One person’s catalysts projected onto another person’s accumulated writing surfaces where their thinking overlaps and where it diverges.

The agent knows about applied targets

When you’ve run apply, Enzyme remembers. The petri output includes applied targets with their path, document count, and when they were applied. The agent sees this and can offer to search the external corpus when a theme bridges both:

“Your catalysts have been applied to research-papers (189 docs) — want me to search there too?”

The agent makes this offer when the current thread plausibly spans the external content. It stays quiet about applied targets when the conversation is clearly internal to the workspace.