Supported runtimes
Run enzyme install <target> from the markdown workspace you want the agent to use. This installs a skill to help with Enzyme setup and writes runtime instructions so the agent knows how to run the retrieval loop.
Runtime matrix
| Runtime | Command | Status | What gets written | How to verify |
|---|---|---|---|---|
| Codex and AGENTS.md-aware agents | enzyme install codex | Supported | Enzyme section in AGENTS.md; .agents/skills/enzyme/SKILL.md | Start the agent from the workspace and ask: Set up Enzyme for this vault. |
| Claude Code | enzyme install claude | Supported | Enzyme section in AGENTS.md; .claude/skills/enzyme/SKILL.md; CLAUDE.md import of @AGENTS.md | Start Claude Code from the workspace and ask: Set up Enzyme for this vault. |
Pi / generic .agents readers | enzyme install codex | Supported | Same as Codex: AGENTS.md plus .agents/skills/enzyme/SKILL.md | Start Pi from the workspace and ask it to read the Enzyme skill before setup. |
| Hermes | enzyme install hermes | Experimental | Workspace instructions plus global ~/.hermes/skills/enzyme/SKILL.md | Start Hermes from the workspace and ask it to set up Enzyme. Expect occasional integration caveats. |
| OpenClaw | enzyme install openclaw | Experimental | OpenClaw-compatible instructions plus global ~/.openclaw/skills/enzyme/SKILL.md | Start OpenClaw from the workspace and ask it to set up Enzyme. Expect occasional integration caveats. |
| Cursor | Manual / pending | Not a dedicated install target | Use a generic AGENTS.md path only if your Cursor workflow reads those instructions. | Treat as manual until a dedicated setup page exists. |
| Product SDK | Private beta | Not part of local runtime setup | Product-corpus integration is scoped directly with early teams. | Use the SDK preview page or contact Enzyme. |
Recommended first target
If you are unsure, use:
enzyme install codex
That writes the generic .agents skill path many coding-agent workflows can read. If you use Claude Code, use enzyme install claude so Claude gets its dedicated skill path too.
What the agent should do next
After runtime install, ask the agent:
Set up Enzyme for this vault. First run a read-only scan, explain what is indexable or weakly indexable, then initialize and prove it with a project or meeting-prep query that cites source files.
A good agent-led setup should preserve your existing markdown conventions. It should not invent a new memory architecture, rewrite your vault wholesale, or write generic summaries. It should also respect provider intent: default init/refresh use Enzyme hosted credits/auth, and agents should pass --use-env-llm only when you intentionally want your own OpenAI/OpenRouter/OpenAI-compatible provider.