Published: 2026-05-30

Give Hermes Permanent Memory with a Free Obsidian Vault

Chapters / key moments (click to jump — plays here on the page)

Without a persistent memory layer, Hermes starts from scratch on every session — it doesn't know your goals, your past work, or your context. Wiring it to an Obsidian vault solves this completely. The vault is just a folder of markdown files, which means Hermes, Claude, and OpenClaw can all read from the same source of truth. Hermes also writes new notes back automatically, so the memory grows without any manual effort.

Source video

"Hermes Obsidian is Insane (FREE!)" by Julian Goldie SEO — Watch on YouTube →

Key Takeaways

Point Hermes at a local Obsidian vault once — every subsequent task starts with full context: your goals, clients, past work, and brand voice.
The vault is plain markdown files, so Hermes, Claude, and OpenClaw can all read the same folder. All agents speak with a consistent, personalized voice.
Hermes writes new notes back to the vault after each session automatically, so context accumulates without you having to maintain it manually.
Responses become grounded in your actual situation rather than generic, reducing the need to re-explain context on every conversation.
Messy vaults aren't a problem — you can ask Hermes to organize, tag, and clean up your notes as a one-time task.
Both Hermes and Obsidian are free and open source; this is a zero-cost upgrade to your agent setup.

Why This Matters: The Context Re-Paste Problem

Every AI agent — Hermes, Claude, ChatGPT — starts each session knowing nothing about you. The standard workaround is to paste a context block at the start of every conversation: who you are, what you're working on, what your goals are. This is tedious, error-prone, and means you're always one forgotten paste away from generic output.

The Obsidian vault approach solves this at the architecture level. Instead of pasting context, you have a local folder that your agents read on every run. The first time is setup; every subsequent session is instant. And because Hermes can update the vault itself, your context document gets richer over time automatically — it notes what you built last week, what you decided, what you want to try next.

Sharing One Brain Across Multiple Agents

The most powerful aspect of this setup is that the vault isn't tied to Hermes specifically — it's a shared memory layer for your entire agent stack. If Hermes completes a task on Monday and Claude answers a question on Tuesday, they both read from the same vault. They reach the same conclusions about your context, use the same voice, and reference the same history. This eliminates the fragmentation that comes from each agent living in its own silo with no awareness of what the others have done.

Related on OpenClawDatabase

Hermes Memory Architecture — how persistent memory works in depth
Hermes Agent Hub — overview, setup, and all guides
OpenClaw SOUL.md — the equivalent persistent context layer for OpenClaw
Compare agent platforms — Hermes vs OpenClaw vs Claude Cowork

What you can actually set up from this

Extracted from the video's own transcript — the specifics the original summary left out.

Reproducible steps

Treat the vault as plain Markdown
An Obsidian vault is just folders of Markdown files, which any agent can read. Point Hermes (and Claude, OpenClaw etc.) at the same folder so they share one set of notes.
Save the vault path with the memory tool — and check it landed
Ask explicitly: "use the memory tool to save the vault path /path/to/vault", then confirm the entry exists with cat ~/.hermes/memories/MEMORY.md. The Hermes docs list this exact case as the most common memory failure.
Let the agent write notes back
His loop: every job drops a note into the vault, so the next agent (or session) starts with it. He also syncs Omi voice-note memories into the vault.

Gotchas

The video says "just tell Hermes where your vault is". The Hermes docs warn that the agent can reply "Done, I'll remember that" without calling the memory tool — the path is then gone next session. Small local models (under ~30B) do this often.
Memory is per profile: a bot on another profile (hermes -p work) reads ~/.hermes/profiles/work/memories/, not the default profile's notes.
Hermes also has a bundled obsidian skill that cron jobs can attach with --skill obsidian (per the docs' arXiv-to-Obsidian example).