How I Set Up The Best Hermes Agent Memory (Free)

Q: Where does Hermes Agent store its memory?

The official docs say the built-in memory lives in two files, MEMORY.md and USER.md, inside ~/.hermes/memories/. A named profile keeps its own copy inside ~/.hermes/profiles/ /memories/.

Q: Why did Hermes forget something I told it to remember?

The most common cause is that the model said it saved the fact without calling the memory tool. Open the memory file to check, ask it to use the memory tool by name, and start a new session, because the memory snapshot is frozen when a session starts.

Q: Can I use more than one Hermes memory provider?

Only one external memory provider can be active at a time, according to the official docs. The built-in memory stays active alongside whichever provider you choose.

Q: Is Hermes Agent memory private?

The built-in memory and an Obsidian vault are plain files on your own machine. A cloud memory provider stores your data on its servers, so read its terms first and keep passwords and API keys out of every memory store.

The best Hermes Agent memory setup is the built-in memory with an Obsidian vault on top as a shared second brain, and that is the exact setup I run every day.

The built-in memory is always on and costs nothing, and the vault gives Hermes somewhere much bigger to read from and write to.

Below that, I rank per-profile memory third, the local Holographic provider fourth and Honcho fifth.

This is the operator version, so you get the ranked list first and then the setup steps for each option.

I'll also show you the privacy mistake I nearly made on camera, because memory is the one part of Hermes that can leak your personal details.

The Top 3 Hermes Agent Memory Picks At A Glance

🥇 Built-in memory plus an Obsidian vault is my number one, because it is free, it stays on your machine and every agent I run can share it.
🥈 Built-in memory on its own is the best starting point, because it is already switched on and needs no setup at all.
🥉 Per-profile memory is the best way to keep separate clients or jobs apart, because each profile gets its own memory files.

How I Ranked The Best Hermes Agent Memory Options

I ranked these on four things.

I ranked on whether I actually use it, what it costs, where your data ends up and how much setup it takes.

My top three are the ones I run daily.

I ranked the external providers from the official Hermes docs and from how they fit a real workflow, not from months of testing each one.

How Hermes Agent's Built-In Memory Works

You need to understand the built-in memory first, because every other option sits on top of it.

The official memory docs say that two files make up the agent's memory.

File	What it holds	Size limit
MEMORY.md	It holds the agent's own notes, such as facts about your environment, conventions and lessons learned.	The limit is 2,200 characters, which is about 800 tokens.
USER.md	It holds your profile, such as your preferences and how you like to communicate.	The limit is 1,375 characters, which is about 500 tokens.

Both files live in the ~/.hermes/memories/ folder.

Hermes loads them into the system prompt as a frozen snapshot when a session starts.

That frozen snapshot means a fact saved in this session only shows up in the next one.

The agent manages the files itself with a memory tool that can add, replace or remove an entry.

Memory does not compact itself, so when a file is full the tool returns an error and the agent has to merge or remove entries to make room.

There is also a second layer called session search.

The docs say every CLI and messaging session is stored in a local SQLite database with full-text search, so Hermes can look up a conversation from weeks ago.

The 7 Best Hermes Agent Memory Options, Ranked

1. Built-In Memory Plus An Obsidian Vault

This is a normal folder of plain text notes that Hermes reads before it works and writes to after it finishes.

I use Obsidian to hold the folder, because it is free and it keeps every file on my own machine.

It is best for anyone who runs more than one agent or wants memory that survives a model change.

It costs nothing.

You get it by making a vault and giving Hermes the full folder path, which I walk through below.

2. Built-In Memory On Its Own

This is the MEMORY.md and USER.md pair plus session search, exactly as Hermes ships.

It is best for beginners and for anyone running one agent on one machine.

It costs nothing, and it is active from your first session.

3. Per-Profile Memory

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

It is best for keeping separate clients, companies or roles from mixing their notes.

It costs nothing, and you create one with the hermes profile create command.

4. Holographic

Holographic is a local SQLite fact store with full-text search and trust scoring.

The docs list it as the local-only choice that needs no external service.

It is best for operators who want more recall than two small files without sending data anywhere.

It costs nothing, and you select it with hermes memory setup.

5. Honcho

Honcho is an external provider maintained by Plastic Labs that builds a model of you across sessions.

The docs describe it as the best fit for multi-agent systems with cross-session context.

It runs on Honcho Cloud at Honcho's own pricing, or you can self-host it for free.

You get it with hermes plugins install honcho and then hermes memory setup.

6. The Other Official Providers

The docs also name Mem0, Hindsight, Supermemory, OpenViking, ByteRover, RetainDB and Memori.

Each one suits a different need, most have a free self-hosted or local mode, and I cover them in a table further down.

7. Memory Helper Plugins From The Catalog

These plugins are not memory stores, but they make the built-in memory safer and easier to see.

They are free, and you find them by running hermes plugins search memory.

Best Hermes Agent Memory Options Compared

Rank	Option	Where the data lives	Cost
1	Built-in memory plus an Obsidian vault	It lives in files on your own machine.	It is free.
2	Built-in memory only	It lives in two files in your Hermes folder.	It is free.
3	Per-profile memory	It lives in each profile's own folder.	It is free.
4	Holographic	It lives in a local SQLite database.	It is free.
5	Honcho	It lives on Honcho Cloud or your own server.	It is paid on the cloud and free when self-hosted.
6	Other official providers	It depends on the provider you pick.	Most have a free mode, and RetainDB is listed at $20 a month.
7	Helper plugins	They read and protect your existing memory files.	They are free.

🔥 Want my exact memory vault template for Hermes? Inside the AI Profit Boardroom, I've got step-by-step videos on wiring Obsidian into Hermes as a shared second brain, plus the installable Agent OS. You also get four coaching calls a week with 3,400+ members. → Get access here

How To Set Up The Built-In Memory Properly

The built-in memory works out of the box, but four small habits make it far more reliable.

Step 1: Check That A Save Really Happened

Memory only persists when the model calls the memory tool.

A reply like "Done, I'll remember that" is only text, and the docs call this the most common reason Hermes seems to forget.

You can open the file and look for the entry yourself.

cat ~/.hermes/memories/MEMORY.md
cat ~/.hermes/memories/USER.md

Only do that when nobody is watching your screen, which I'll explain in the privacy section.

Step 2: Ask For The Tool By Name

If the fact is missing, tell Hermes to use the memory tool to save it.

The docs say small local models and models with weak tool calling often claim a save they never made.

The fix they give is a stronger model for the setup, because a smaller model can still read entries once they exist.

My post on the best Hermes Agent model covers which brains handle tools well.

Step 3: Start New Sessions At Natural Breaks

A Telegram or Discord chat is one long session, and new memory entries stay invisible to it because of the frozen snapshot.

The docs recommend running /new when you finish a task, change topic or start a new day.

Step 4: Turn On Write Approval If You Want Control

By default, Hermes saves memory freely, including from a background review that runs after a turn.

You can set memory.write_approval: true in your config so every save waits for your yes or no.

You then review saves with /memory pending and apply them with /memory approve.

That is the official answer to an agent that saved a wrong assumption about you.

How To Set Up An Obsidian Vault As Hermes Memory

This is my real setup, and the full walkthrough is in my Hermes second brain guide.

Step 1: Create The Vault

Install Obsidian and create a new, empty vault.

A vault is only a folder, so there is no database and no lock-in.

Step 2: Give Hermes The Full Folder Path

Tell Hermes the actual path, such as ~/SecondBrain, and ask it to build the folders inside it.

I once said "this folder" instead of the full path, and Hermes built everything in its own workspace instead.

Step 3: Build Nine Folders

I use seven folders from a system called PARA, plus two more that turn it into an agent brain.

The seven are Inbox, Daily, Projects, Areas, Resources, Memories and Archive, and the extra two are a Wiki folder and an Agentic OS folder with one sub-folder per agent.

Step 4: Write Three Anchor Notes

Write an About You note, keep your live projects and areas current, and write a Goals note, because an agent with no goals optimises for nothing in particular.

Step 5: Add A Protocol File

One file tells every agent where things live, what to read at the start and what to write at the end.

My whole protocol comes down to one rule, which is to read before you work and write before you leave.

It also tells agents to append and never wipe, so no agent overwrites another agent's notes.

Step 6: Add The Context Rules

My protocol has five context rules, and the fifth one does most of the work.

It tells the agent never to dump the whole vault, to pull only what the question needs and to say so when the answer isn't there.

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

That gap is the reason the clean wiki layer exists.

Step 7: Tell Hermes Where The Vault Lives

Ask Hermes to use the memory tool to save the vault path.

The docs add that a skill is often a better home for a location a recurring task needs, and my post on the best Hermes Agent skills covers that.

The same vault works with Claude, and my Claude Obsidian second brain setup shows that side of it.

🚀 Ready to put How I Set Up The Best Hermes Agent Memory to work in your business? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom

How To Set Up Per-Profile Memory

Profiles are how I'd separate clients, and the profiles docs are clear that memory is scoped per profile by design.

You create a profile with one command.

hermes profile create coder --clone

The --clone flag copies your config, your SOUL.md, your skills and the two curated memory files, and after that the profile uses its own folder at ~/.hermes/profiles/coder/memories/.

A CLI session on the default profile and a Telegram bot on another profile do not share notes.

The docs give one hard warning here.

Never point two agent processes at the same Hermes home, because both write memory automatically and they will compound each other's entries.

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

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

How To Set Up An External Memory Provider

Every provider uses the same three commands.

hermes memory setup     # pick a provider and configure it
hermes memory status    # see what is active
hermes memory off       # go back to built-in memory only

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

The built-in memory keeps working exactly as before, because the provider is additive.

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

Here is how the docs describe each provider.

Provider	What the docs say it is best for	Cost in the docs
Holographic	It suits local-only memory with no external dependencies.	It is free.
Honcho	It suits multi-agent systems with cross-session context.	It is paid on the cloud and free when self-hosted.
Mem0	It suits hands-off memory, because Mem0 extracts facts automatically.	It is paid on the platform and free when self-hosted.
Hindsight	It suits knowledge-graph recall with entity relationships.	It is paid on the cloud and free locally.
Supermemory	It suits semantic recall with user profiling.	It is paid on the cloud and free when self-hosted.
OpenViking	It suits self-hosted knowledge management with structured browsing.	It is free and open source.
ByteRover	It suits developers who want portable, local-first memory with a CLI.	It is free locally.
RetainDB	It suits teams already using RetainDB.	It is listed at $20 a month.
Memori	It suits agent-controlled recall with project and session attribution.	It follows Memori's pricing.

Which providers come bundled depends on your Hermes version.

On my install, hermes memory --help lists Honcho, OpenViking, Mem0, Holographic, RetainDB and ByteRover as bundled.

The newest docs say several of those have moved to the plugin catalog, so run the help command on your own machine before you follow any tutorial.

Memory Plugins Worth Knowing About

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

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

These four helper plugins caught my eye, and the descriptions come straight from the catalog.

hermes-memory-wiki is the one marked official, and it adds a Memory Wiki dashboard tab with a read-only audit panel for MEMORY.md and USER.md.
memory-rewind keeps an automatic version history of the built-in memory so you can diff and restore past versions.
memory-shield stops 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 other people's code, so read each plugin's disclosure before you install it.

My guide to the Hermes Agent plugins marketplace explains how the catalog and its tiers work.

🔥 Want me to look at your Hermes memory setup live? Inside the AI Profit Boardroom, you can share your screen on one of four coaching calls a week and get your vault, profiles and memory rules fixed. There are 3,400+ members building the same stack. → Join the AI Profit Boardroom

The Privacy Mistake I Nearly Made With Hermes Memory

When I filmed my Herald OS video, I opened the Memory page to show everything Hermes remembered.

My Telegram details were sitting right there on screen.

Memory is a list of true facts about you, so treat every memory page like an open password manager.

Here are the rules I follow now.

I check what's in memory before I share my screen, record a video or join a call.
I never store passwords, API keys or tokens in memory or in the vault.
I read a cloud provider's data terms before I send it a single conversation.
I turn on write approval on any agent that talks to other people in group chats.

Hermes does help here.

The docs say memory entries are scanned for injection and exfiltration patterns before they are accepted.

They also say everything handed to an external provider passes through a secret scrub first.

Those are safety nets, and they don't replace the habit of keeping secrets out.

You can read more about that interface in my Herald OS walkthrough.

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.

Store this	Leave this out
Store your preferences, such as the tone and format you like.	Leave out passwords, API keys and tokens.
Store environment facts, such as your operating system and tools.	Leave out facts the agent can find again with a quick search.
Store corrections you had to make more than once.	Leave out raw data dumps such as logs and big code blocks.
Store project conventions and where things live.	Leave out temporary file paths and one-off debugging context.
Store completed work with a date.	Leave out anything already written in your SOUL.md or AGENTS.md files.