Hermes

Experimental

Hermes support is experimental. Enzyme connects Hermes to the markdown workspace where your durable knowledge already lives. Instead of waiting for memory to accumulate through chat, Hermes can start from notes, decisions, meetings, people pages, project logs, and session summaries you already have.

Markdown stays the source of truth. Enzyme indexes the vault locally, preserves the structure already present, and refreshes new markdown as the agent writes.

What changes

With Enzyme active, Hermes gets:

  • Vault recall before the answer: relevant notes can be retrieved before Hermes responds.
  • Continuity across sessions: current questions can be connected to older work, decisions, people, and projects.
  • Markdown writeback: durable session facts can be written into the vault and refreshed into future recall.

Hermes still has its own session history. Enzyme gives it a structured view of the workspace around that session.

Why this is different

Most memory providers start from conversation history and derive user facts over time. Enzyme starts from the user’s existing corpus. It reads folders, dates, tags, frontmatter fields, and wikilinks as retrieval signal.

That matters for Obsidian users and teams with established knowledge bases. Setup should recognize patterns like:

  • inbox/, daily/, meetings/, projects/, people/, clients/, or similar folders
  • frontmatter fields such as tags, created, date, people, companies, clients, and projects
  • wikilinks for people, organizations, ideas, and recurring workstreams

Enzyme can suggest improvements, but it should not silently reorganize a vault.

Install Enzyme

Install the Enzyme CLI on the Hermes machine:

curl -fsSL enzyme.garden/install.sh | bash

Then install the Hermes skill from the workspace root you want indexed:

cd /path/to/workspace
enzyme install hermes
hermes

The Quickstart covers local CLI setup. This page covers the Hermes memory path.

Configure Hermes memory

Start Hermes from the workspace root after running enzyme install hermes. The installed skill gives Hermes the setup and retrieval workflow for that markdown corpus.

First run

In a new Hermes session, ask:

Set up Enzyme for this workspace.

Hermes should get Enzyme ready and then use it for workspace recall. For the full agent workflow, see the Enzyme agent skill.

Available commands

Hermes can use the normal Enzyme CLI from the workspace root:

CommandWhat it does
enzyme scanInspects markdown folders, frontmatter, tags, and likely vault structure.
enzyme initBuilds the index, catalysts, embeddings, and similarities.
enzyme petriReturns active entities, catalysts, and orientation context.
enzyme catalyzeSearches through catalyst-guided semantic recall with file/date/source traces.
enzyme refreshUpdates the index after new markdown writes.

The Enzyme skill is the durable instruction set for setup, retrieval, and markdown writeback policy.

Verify the integration

Ask Hermes:

What has been active across this workspace lately, and what older notes should inform what I work on next?

A useful answer should cite concrete files and connect themes across time. If it only lists filenames or recent edits, the config probably missed the folders where the real work lives.

For a fuller test plan, see Verify memory.

Configuration

Hermes uses the workspace where it is launched. Run enzyme install hermes from that markdown workspace so Enzyme can write the workspace marker and install the runtime skill.

Vault-specific Enzyme settings live in ~/.enzyme/config.toml under the workspace’s absolute path. The important fields are the selected entities, excluded noisy folders/tags/links, and optional targets. enzyme init and enzyme refresh use Enzyme hosted credits/auth by default; provider environment variables such as OPENAI_API_KEY are used only when you intentionally run with --use-env-llm.

These settings should describe the vault the user already has. They should not force a new memory hierarchy.

Next steps