Zhar Consulting

Knowledge Base - Part 1 - Setup

Properties2
tagsai, llm, local_llm, tools
aliasesKnowledge Base Setup, Obsidian and Hermes setup

3 min read

This is Part 1 of building a local knowledge base on top of my Obsidian vault, using the Hermes agent to process it. This part covers the setup on macOS: installing Obsidian, installing Hermes, and wiring Hermes up to a GLM coding plan key from Z.ai. Part 2 covers turning the vault into a knowledge graph with graphify.

Why these two?

Obsidian is where the knowledge lives, as plain Markdown files. Hermes is the open-source, self-hosted AI agent by Nous Research that does the processing. Since everything is local files, the two compose well: the agent reads and writes the same vault I edit by hand.

Install Obsidian

  1. Download the installer from obsidian.md/download and drag it to Applications. Or with Homebrew:
brew install --cask obsidian
  1. Open Obsidian and create a vault — this is just a folder of .md files on disk. Mine lives at ~/Documents/obsidian-vault.
  2. If you are new to Obsidian, the official help docs are excellent, and the vault itself is fully documented if you ever want to script around it.

That is the whole install. No account needed.

Install Hermes

Docs: https://hermes-agent.nousresearch.com/docs/

The installer handles everything (Python 3.11 via uv, Node.js v22, ripgrep, ffmpeg, the repo clone, a venv, and the global hermes command). The only hard prerequisite on macOS is git — check with git --version.

Option A — Desktop app (recommended): Download the Hermes Desktop installer and run it. This installs both the desktop app and the CLI.

Option B — Command line only:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

Add the desktop app later any time with hermes desktop.

Verify the install:

source ~/.zshrc
hermes doctor    # diagnostics — reports anything missing and how to fix it
hermes           # start chatting

If you get hermes: command not found, reload your shell or check that ~/.local/bin is on your PATH. Everything Hermes-related lives under ~/.hermes/.

Set up Hermes with a GLM coding plan key

Instead of pay-as-you-go API keys, Z.ai offers the GLM Coding Plan — a flat monthly subscription that works with coding agents through an OpenAI-compatible endpoint. This is the cheapest way to run Hermes on a frontier model. (I originally used a Mistral coding plan token, but Mistral’s Pro plan does not work with third-party agents like Hermes, while the GLM plan does.)

Docs: https://docs.z.ai/devpack/overview

  1. Subscribe to the plan: log in at z.ai and pick a GLM Coding Plan — Lite, Pro or Max, starting at $18/month.
  2. Create an API key under Plan Overview. Treat it like a password — it is billed per use.
  3. Point Hermes at Z.ai. Hermes speaks any custom OpenAI-compatible endpoint via the model picker:
hermes model

Choose the custom OpenAI-compatible endpoint option and enter:

  • Base URL: https://api.z.ai/api/coding/paas/v4
  • API key: your GLM coding plan key
  • Model: glm-5.3 (or glm-5.3-flash for the cheaper, faster variant — see the model docs)

Keys are stored in ~/.hermes/.env. You can also set them directly:

hermes config set ZAI_API_KEY <your-key>
  1. Check what is active:
hermes config get model

Sanity check

Run a quick chat in your vault folder to confirm the model answers:

cd ~/Documents/obsidian-vault
hermes

Ask it something trivial, and once it responds you are done with the setup. Continue with Part 2 — Graphify to turn the vault into a queryable knowledge graph.

Graph View

Backlinks