Docs
Evaluate
Decide whether Enzyme fits your workspace
Enzyme makes your existing markdown workspace operable by agents, avoiding a hosted memory database.
Try
Installation, trust, and repair
Get from install to a context layer for your agent, and understand the local footprint.
Quickstart
Install the CLI, customize the runtime instructions, and initialize Enzyme on your markdown workspace.
Supported runtimes
Claude, Codex, Pi, and other harnesses.
Data flow & trust
Local storage, provider dependencies, and cleaning up
Troubleshooting
Generic output, provider errors, stale indexes, and noisy results
Understand
Learn the retrieval model
After you see it work, learn more about Enzyme's primitives, and why Enzyme is not just search.
How it works
The compile loop: scan → init → petri → catalyze → refresh.
Catalysts
Generated questions that connect source files keyword and vector search miss.
Memory is not search
When exact search is better, and when agent memory needs source-grounded context.
What the agent sees
How structural understanding changes the first turn of a session.
Use
Make the trail durable
Keep memory inspectable by writing conservative markdown artifacts and refreshing them into future retrieval.
Capture memory as artifacts
Append logs, decisions, observations, and reviewable writeback boundaries.
Verify memory
Test recall, source citations, writeback, refresh, and cross-session continuity.
Apply
Project your workspace lens onto an unfamiliar corpus, with caveats.
In practice
Short clips from real sessions where Enzyme changed what the agent found.
Evidence & compare
Place Enzyme against alternatives
Use speed claims carefully, compare by job-to-be-done, and choose another tool when it fits better.
Benchmarks
What the 8ms query-time retrieval claim measures, and what it does not measure.
Comparisons
Enzyme vs exact search, RAG, managed memory APIs, coding-agent memory, and larger context.
Configuration
Entity selection, embedding limits, provider setup, and config.toml.
SDK preview
Private-beta product-corpus path, separated from the local CLI flow.