About Enzyme
Compile-time intelligence for knowledge bases. Enzyme grew out of recommendation systems that had to take imprecise, personal, half-formed signals seriously — and turn them into something an agent could act on.
From taste to knowledge
Enzyme's roots are in agentic playlist curation at Spotify — systems that had to understand what “Sunday morning cleaning music” meant when the user couldn't quite articulate it themselves. A mood and a tempo and a context, compressed into five words. That work shipped to hundreds of millions of users.
Enzyme is the same problem from a different direction. Instead of understanding taste from listening behavior, it extracts conceptual structure from accumulated writing — notes, highlights, transcripts, agent outputs. The engine compiles a knowledge base into a concept graph in under 15 seconds, then serves semantic queries in 8ms on device. No conversation history needed. No cold start.
The problem
Every knowledge base has the same gap. The content is there — documents, highlights, transcripts, saves — but the conceptual structure is implicit. An agent can search by keyword. It can't see the recurring tensions, the themes that span months, the connections between captures that don't share vocabulary.
Most memory tools try to build this understanding at runtime, through conversation. They start empty and accumulate. That works when the primary interaction is chat. It doesn't work when the content already exists and the intelligence layer needs to be ready on day one.
The approach
Enzyme treats understanding as a compile step. It reads the structure of a corpus — tags, links, folders, timestamps — and generates catalysts: thematic questions that cut across the content and name what's latent in it. Those catalysts become the search layer. Queries resolve through pre-computed relationships instead of runtime inference. The research page covers how this compares to query-time memory systems.
The engine is built for corpora that grew before anyone planned them. Half-finished tags, abandoned folders, links that meant something at 2am — these are real signals, not noise. A knowledge base doesn't need to be organized before Enzyme can read it. The conceptual shape is already there.
Two products, one engine
For individuals — a local CLI and Claude Code plugin that indexes Obsidian vaults, Readwise exports, and any markdown corpus. 42,000+ downloads. Everything runs on device.
For product teams — the same compile-time engine as an SDK for applications where users bring accumulated content: reading highlights, saved recipes, design explorations, annotated research. The concept graph is a portable artifact — compile it from what the user imports, and every agent session starts with understanding rather than building it from scratch. Learn more.
Get in touch
Building a product on accumulated user content? We scope the corpus, refresh model, and deployment path together.
About the founder
Joshua Pham is an ML engineer based in New York. He built agentic recommendation systems at Spotify serving hundreds of millions of users, started Enzyme in 2023 as a tool for his own research practice, and left Spotify in 2025 to work on it full-time. By then it had been his daily driver for two years — one of the first MCPs, integrated with Claude Code since early 2025.
Founding members are shaping what comes next in Discord.