Choosing a Hermes memory provider is the one storage decision that outlives every model you’ll ever run. Here’s what the provider layer actually is, the options compared, and why the readable one wins.
Short answer
- The provider is the backend where persistent memory lives — swappable since the modular-memory era.
- Options: default file store, flat folder, an Obsidian vault, or a vector DB.
- The vault wins because you can read it — unreadable memory is uncorrectable memory.
- One provider can serve every agent you run, and it exports in one file.
What a Hermes memory provider actually is
Hermes separates having memory from where that memory lives. The provider is the backend — the thing that actually stores and serves what your agent remembers — and since the v0.7 era made the memory system modular, it’s a swappable part rather than a fixed one.
That separation is the whole point. Your agent’s knowledge isn’t welded to one storage mechanism, one machine or one model. Change the provider and the agent keeps being the agent.
The Hermes memory provider options
| Provider | What it is | Verdict |
|---|---|---|
| Default file store | Plain markdown files Hermes manages itself | Fine to start — zero setup |
| Obsidian vault | A linked markdown vault your agents read and write | The one I run — human-readable, linkable, shared across agents |
| Flat notes folder | Text files without linking | Works, degrades as it grows |
| Vector database | Embeddings and semantic search | Powerful at scale, unreadable by you |
The reason the vault wins is boring and decisive: you can open the files. Every memory is a markdown note you can read, correct by editing text, and audit. A provider you can’t read is a provider you can’t trust — full reasoning in the best-memory breakdown.
🔥 Want this set up without the guesswork? Choosing and wiring the memory provider is a one-time job that decides how smart every future session starts. Inside the AI Profit Boardroom you get the Agent OS as a ready-to-install file, a 30-day roadmap, daily tutorials the same day new tools ship, and four live coaching calls a week where you share your screen and get unstuck. → Get access here
Why a swappable memory provider matters
- Models change; memory shouldn’t. Swap the brain and the provider carries your context to the new one untouched.
- Every agent can read the same provider. Point Claude, Hermes and OpenClaw at one vault and what one learns, all know.
- It travels. Export packages the memory with the agent — credentials stripped — so the provider’s contents move machines in one file.
- Per-profile isolation still works. Each profile keeps its own memory space, so a shared provider doesn’t mean cross-contamination.
Setting your provider up
- Start with the default store — confirm the agent reads and writes memory at all.
- Move to an Obsidian vault once it matters: point Hermes at the vault folder and seed a handful of notes.
- Let the agents maintain it — they write markdown; you don’t.
- Verify with a recall question about last week. If the answer cites real specifics, the provider is live.
- Back it up with export — the provider’s contents are the one part of your setup you can’t rebuild.
Full wiring in the memory setup guide, and the wider architecture in how the memory system fits together.
FAQ
What is a Hermes memory provider?
The backend where your agent’s persistent memory actually lives. Hermes made memory modular, so the provider is a swappable part — default file store, an Obsidian vault, or something heavier.
Which memory provider is best?
An Obsidian vault of plain markdown — human-readable, linkable, shared across every agent you run, and correctable by editing a text file.
Can I change provider later?
Yes — that’s the point of the modular design. The memory is markdown, so moving it is moving files, not migrating a database.
Do all my agents share one provider?
They can, and should: one vault readable by Hermes, Claude and OpenClaw means context is learned once, not per tool.
Is a vector database better?
At serious scale, sometimes — but you can’t read it yourself. Start with markdown you can audit and add retrieval only if you genuinely outgrow it.
How do I back the provider up?
Run export — it packages memory with persona, skills and schedules into one file with credentials stripped automatically.
The bottom line on the Hermes memory provider
The Hermes memory provider is the swappable backend behind everything your agent remembers — and the right choice is the one you can read. Run a markdown vault, share it across your agents, let them maintain it, and export it weekly. Brains come and go; the provider is the part you keep.
About Julian Goldie
I run Goldie Agency, a 7-figure SEO agency, and teach this daily on a 400K+ subscriber YouTube channel. 240+ client projects on Upwork at a 100% job-success score, 10+ years through every major Google update. My systems are in the AI Profit Boardroom; my link building book is free here.
Related reading
Last updated August 2026. This is the living guide to Hermes memory provider — it gets updated as the tools change.
