agentos.guide › blog

The Best Hermes Agent Memory, And How It Becomes The Memory Layer Of Your Agent OS

By Julian Goldie · 10 October 2026 · agentos.guide

The best Hermes Agent memory is the built-in memory with an Obsidian vault on top, because it is free, it stays on your machine and every agent in your Agent OS can share it.

The built-in memory on its own is my second pick, and per-profile memory is my third.

This is my own ranking, based on the setup I run every day and on the official Hermes docs.

Here is the ranked list first, then a six-step quick-start, then how memory becomes the third layer of an Agent OS.

The best Hermes Agent memory options, ranked for an Agent OS

Here is the full ranking, with who each option suits, what it costs and where your data ends up.

I put the vault first because it is the only option on this list that every agent in my Agent OS can share for free.

The built-in memory is second because it is always on, and you should never switch it off.

Per-profile memory is third because it is how you keep separate jobs from mixing their notes.

The top three at a glance

• 🥇 Built-in memory plus an Obsidian vault is the best overall, because it is free, private and shared.
• 🥈 Built-in memory on its own is the best starting point, because it needs no setup.
• 🥉 Per-profile memory is the best way to give each agent or client a separate memory.

How I ranked them

I ranked on whether I use it daily, what it costs, where the data lives and how much setup it needs.

My top three are what I run every day.

I ranked the external providers from the official Hermes docs, not from months of testing each one, so treat that part as a map.

How Hermes Agent's built-in memory works

Hermes Agent is the free, open-source agent from Nous Research, and it ships with a small memory that persists between sessions.

The official memory docs describe it like this.

Together the two files come to roughly 1,300 tokens, which is a few short paragraphs.

That is deliberate, because those facts are paid for in every single prompt.

The frozen snapshot means a fact saved in this session shows up in the next session, not the current one.

Session search covers the rest, so Hermes can still find a conversation from weeks ago when it needs the detail.

Quick-start: the best Hermes Agent memory in six steps

This is the short version of my own setup, and it takes about an afternoon to do properly.

Give Hermes the actual path, such as ~/SecondBrain, and not the words "this folder".

I made that exact mistake while filming, and Hermes built the folders in its own workspace instead.

The full walkthrough, with the real protocol file, is in my Hermes and Obsidian second brain guide.

🔥 Want the exact Hermes Agent memory in the Agent OS setup I run inside my Agent OS?

Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video tutorials, four coaching calls a week and 3,400+ members building real automations.

→ Get access here

Where memory fits in an Agent OS

I think of an Agent OS as three layers.

The first layer is the brain, which is Hermes and whichever model you plug into it.

The second layer is the cockpit, which is the dashboard you drive it from.

The third layer is the memory, which is a notes vault your agents read and write.

Without the third layer, every session is a first date.

The memory layer is also the one that survives everything else.

When a new model comes out, I change a dropdown, and the new brain inherits the same vault.

The 3-Layer Hermes OS guide covers how the three layers fit together.

How the vault works as a shared second brain

A second brain is a folder of text files, and people make it far more complicated than it needs to be.

Obsidian is the free app that shows those files and keeps them on your own machine.

Because the notes are only text, every agent I run can read the same brain.

Seven of my agents read one vault, which are Hermes, Claude, Codex, OpenClaw, Grok, Apollo and Jarvis.

The vault has nine folders, which are Inbox, Daily, Projects, Areas, Resources, Memories, Archive, Wiki and Agentic OS.

The first seven are a system called PARA, and the last two are what make it an agent brain.

A good vault has two layers.

The capture layer is messy, and everything lands there without sorting.

The clean layer is a wiki of short pages that the agent reads first.

When I measured my own vault for the guide, the whole thing was around 870,000 tokens and one clean wiki page was around 122 tokens.

That is why the protocol tells every agent never to dump the whole vault and to pull only what the question needs.

It also tells the agent to say so when the answer isn't in there, which stops it from inventing a client or a number.

The last rule is ownership, which means every agent appends and no agent wipes.

Each agent gets its own folder, so nothing gets overwritten.

Per-profile memory for agent teams

A Hermes profile is a separate Hermes home with its own config, keys, memory, sessions and skills.

The profiles docs say memory is scoped per profile by design.

You create one with hermes profile create coder --clone, and the clone flag copies your two curated memory files across.

After that, a coding profile and a writing profile each build their own notes.

The docs give one hard warning, which is never to point two agent processes at the same Hermes home.

Both would write memory automatically and compound each other's entries.

When two agents need shared memory, the docs say to use an external provider, and I use the vault for the same job.

Herald OS also has Spaces for Ideas, Work and Personal, but I haven't confirmed that a Space keeps a separate memory, so I rely on profiles for real separation.

In an Agent OS, this gives you a clean pattern.

Each profile keeps its private notes, and the shared vault holds what every agent should know.

The external memory providers Hermes officially supports

The memory providers docs name nine providers.

You pick one with hermes memory setup, check it with hermes memory status and switch it off with hermes memory off.

Only one external provider can be active at a time, and the built-in memory stays on alongside it.

When a provider is active, Hermes pulls relevant memories before each turn and syncs the turn back afterwards.

Which providers are bundled depends on your version.

On my install, hermes memory --help lists Honcho, OpenViking, Mem0, Holographic, RetainDB and ByteRover as bundled, while the newest docs say several have moved to the plugin catalog.

Run the help command on your own machine before you follow any tutorial, including this one.

Be careful with hermes memory reset, because it erases all built-in memory rather than switching a provider off.

🔥 Want the exact Hermes Agent memory in the Agent OS setup I run inside my Agent OS?

Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video tutorials, four coaching calls a week and 3,400+ members building real automations.

→ Get access here

Memory plugins from the official catalog

The plugin catalog is separate from the provider list.

I ran hermes plugins search memory on 10 October 2026, and it listed 58 entries in the memory category.

Most of those are community memory providers that I have not tested.

These helpers stood out, and the descriptions come from the catalog itself.

• hermes-memory-wiki is marked official, and it adds a Memory Wiki dashboard tab with a read-only audit panel for the two memory files.
• memory-review lets you tick staged memory writes and approve or reject them in a dialog.
• memory-rewind keeps an automatic version history of built-in memory so you can diff and restore past versions.
• memory-shield keeps the agent from rewriting or wiping what it remembers about you.
• obsidian-memory lets you browse, search and edit Obsidian vault notes from the Hermes dashboard.

Community plugins are third-party code, so read each plugin's disclosure before you install it.

My plugins marketplace quick-start explains how the catalog and its tiers work.

The privacy lesson from my Herald OS video

Memory is a list of true facts about you, so it needs more care than a chat log.

When I toured Herald OS on camera, I opened its Memory page to show everything Hermes remembered.

My Telegram details were sitting right there on screen.

So I now check memory before I share my screen, record a video or join a call.

I keep passwords, API keys and tokens out of memory and out of the vault, and my own protocol file says so in writing.

I read a cloud provider's data terms before I send it anything, because an active provider receives your conversation turns.

The docs say Hermes scans memory entries for injection patterns and scrubs recognisable secrets before handing data to a provider.

Those are backstops, and they don't replace the habit.

If an agent talks to other people, you can set write_approval to true so every save waits for your yes.

You then review saves with /memory pending.

My Herald OS quick-start shows the rest of that interface.

What to store in Hermes memory and what to leave out

The built-in files are tiny, so every entry has to earn its place.

The docs suggest merging entries once a file passes 80% of its limit.

Long material belongs in the vault, where there is no character limit.

The docs also say a skill is often a better home than a memory entry for something a recurring task needs every run.

My ranking of the best Hermes Agent skills covers that side.

Common mistakes with Hermes Agent memory

The first mistake is trusting a reply like "I'll remember that".

Memory only persists when the model calls the memory tool, and the docs say small local models often skip the call.

You can fix it by asking for the memory tool by name and by using a stronger model for setup, which my best Hermes Agent model ranking helps with.

The second mistake is never starting a new session on Telegram or Discord.

A messaging chat is one long session, so new memory entries stay invisible until you run /new.

The third mistake is letting agents overwrite each other in the vault.

An append-only rule and one folder per agent solve that.

The fourth mistake is dumping the whole vault into context, which wastes tokens and still doesn't fit.

The fifth mistake is building the memory layer last.

I'd build it straight after the install, and my best Hermes Agent setup ranking shows where it goes.

How to choose your Hermes Agent memory

Choose the built-in memory if you are new and run one agent.

Choose the vault if you run several agents or want your knowledge to outlive any one model.

Choose profiles if you serve separate clients or run an agent team.

Choose Holographic if you want deeper recall with no cloud.

Choose Honcho if several agents need one shared view of the same user.

Choose another provider only if its niche matches a real problem you have today.

FAQ

What is the best Hermes Agent memory?

My pick is the built-in memory with an Obsidian vault on top. The two built-in files hold the short facts Hermes needs in every session, and the vault holds everything else as plain text that any agent can read.

How does Hermes Agent's built-in memory work?

The official docs say two files, MEMORY.md and USER.md, are stored in the Hermes memories folder and loaded into the system prompt as a frozen snapshot at session start. MEMORY.md is limited to 2,200 characters and USER.md to 1,375 characters.

Which external memory providers does Hermes support?

The official docs name Honcho, OpenViking, Mem0, Hindsight, Holographic, RetainDB, ByteRover, Supermemory and Memori. Only one external provider can be active at a time, and the built-in memory stays on alongside it.

How do I give each agent its own memory?

Create a Hermes profile for each agent or client. The docs say memory is scoped per profile by design, so each profile keeps its own memory files, sessions and keys.

Is Hermes Agent memory safe to show on screen?

Treat it as private. When I opened the Memory page in Herald OS on camera, my Telegram details were visible, so check what is stored before you share your screen and keep passwords and API keys out of every memory store.

Why does Hermes forget what I told it to remember?

The usual cause is that the model said it saved the fact without calling the memory tool. Ask it to use the memory tool by name, confirm the entry exists, and start a new session, because the snapshot is frozen when a session starts.

Final word on the best Hermes Agent memory

Keep the built-in memory on, build a vault this weekend, and give each client or agent its own profile.

Feed the vault a little every day, keep secrets out of it, and you will have the best Hermes Agent memory.

Related guides on agentos.guide

→ Hermes AI + Obsidian: Build The Ultimate AI Second Brain

→ The 3-Layer Hermes OS

→ What Is Herald OS? Free Hermes Agent OS Quick-Start

→ Hermes Agent Plugins Marketplace: 5-Step Quick Start

📺 Video notes + links to the tools 👉

🎥 Learn how I make these videos 👉

🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉

← all guides on agentos.guide