The best Hermes Agent memory is the built-in memory paired with an Obsidian vault, and it gets my top score because it is free, private and shared across every agent I run.
The built-in memory on its own comes second, and per-profile memory comes third.
Hermes Agent gives you more memory choices than most people realise, from two small text files to nine external providers and dozens of plugins.
This verdict scores six options on cost, privacy, recall and setup effort, then goes through each one feature by feature.
The scores are my own first-person ratings, based on what I use daily and on the official Hermes docs, so they are opinions and not lab measurements.
The Verdict On Hermes Agent Memory In 30 Seconds
- 🥇 Built-in memory plus an Obsidian vault wins, because it combines always-on facts with a big, private, shareable knowledge base.
- 🥈 Built-in memory only is the right choice for a beginner, because it works with zero setup.
- 🥉 Per-profile memory is the right choice when you need separate memories for separate jobs.
If you only do one thing after reading this, make a vault and tell Hermes to read it before it works.
Hermes Agent Memory Scores At A Glance
Each score is out of ten, and a higher number is better in every column.
| Rank | Option | Cost | Privacy | Recall depth | Ease of setup | My overall score |
|---|---|---|---|---|---|---|
| 1 | Built-in memory plus an Obsidian vault | 10 | 10 | 9 | 7 | 9.0 |
| 2 | Built-in memory only | 10 | 10 | 5 | 10 | 8.0 |
| 3 | Per-profile memory | 10 | 10 | 5 | 9 | 7.5 |
| 4 | Holographic | 10 | 10 | 7 | 8 | 7.5 |
| 5 | Honcho | 7 | 6 | 8 | 6 | 7.0 |
| 6 | Other official providers | 7 | 6 | 8 | 5 | 6.5 |
The overall score is my judgement and not a simple average, because I weight daily usefulness the most.
How I Scored The Best Hermes Agent Memory Options
Cost is about what you pay each month, and a free option scores ten.
Privacy is about where your data lives, and local files on your own machine score ten.
Recall depth is about how much the agent can actually remember and find.
Ease of setup is about how quickly a normal user gets it working.
I use the first three options every day, so those scores come from experience.
I scored the external providers from the official memory providers docs, so treat those numbers as informed estimates.
🔥 Want my exact Best Hermes Agent Memory setup, step by step? 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
What Hermes Agent Memory Is And How It Works
Hermes Agent is a free, open-source AI agent made by Nous Research.
Its built-in memory is two plain text files that persist between sessions.
| Feature | What the official docs say |
|---|---|
| MEMORY.md | It holds the agent's own notes, and it is limited to 2,200 characters. |
| USER.md | It holds your profile and preferences, and it is limited to 1,375 characters. |
| Location | Both files are stored in the ~/.hermes/memories/ folder. |
| Loading | Both are injected into the system prompt as a frozen snapshot at session start. |
| Editing | The agent uses a memory tool that can add, replace or remove entries. |
| Full memory | Memory does not auto-compact, so the tool returns an error and the agent makes room itself. |
| Session search | Every past session is stored in a local SQLite database with full-text search. |
| Safety scan | Entries are scanned for injection and exfiltration patterns before they are accepted. |
| Write approval | A setting called write_approval makes every save wait for your approval. |
The key idea is that the always-on memory is deliberately tiny.
It holds the facts that must be in every session, and session search finds everything else on demand.
1. Built-In Memory Plus An Obsidian Vault: 9.0 Out Of 10
An Obsidian vault is a folder of plain text notes on your own computer.
Hermes reads the notes it needs before it works and writes a short log after it finishes.
What I like: the vault has no character limit, it costs nothing, and seven of my agents read the same one.
What I don't like: it needs a clear rules file, or agents will overwrite each other's notes.
Best for: this is best for anyone running more than one agent or planning to switch models.
Cost: it is free, because Obsidian is a free app.
How to get it: follow my Hermes second brain guide, which shows the folders and the rules file.
The numbers from my own vault explain why this works.
When I measured it for that guide, the whole vault was around 870,000 tokens, and one clean wiki page was around 122 tokens.
So I tell the agent to read the small clean page first and never to dump the whole vault into context.
2. Built-In Memory Only: 8.0 Out Of 10
This is Hermes exactly as it ships, with MEMORY.md, USER.md and session search.
What I like: there is nothing to install, and the agent saves useful facts on its own.
What I don't like: roughly 1,300 tokens is a few paragraphs, so it can't hold a whole business.
Best for: this is best for a first week with one agent on one machine.
Cost: it is free.
How to get it: you already have it, and you can check it by opening the two files in ~/.hermes/memories/.
One gotcha is worth knowing.
The memory docs say a model can reply "I'll remember that" without calling the memory tool, which means nothing was saved.
The docs add that small local models do this often, so a strong model matters here, and my best Hermes Agent model post covers that choice.
3. Per-Profile Memory: 7.5 Out Of 10
A profile is a separate Hermes home with its own config, keys, memory, sessions and skills.
What I like: it gives clean separation, so a coding agent and a writing agent never mix notes.
What I don't like: the memories are isolated, so sharing a fact between profiles takes a vault or a provider.
Best for: this is best for separate clients, companies or roles.
Cost: it is free.
How to get it: run hermes profile create with a name, and add --clone to copy your current memory files across.
The docs warn that two agent processes must never share one Hermes home, because both write memory automatically.
Herald OS also has Spaces for Ideas, Work and Personal, but I haven't verified that a Space separates memory, so I'd use profiles for real isolation.
4. Holographic: 7.5 Out Of 10
Holographic is a local SQLite fact store with full-text search and trust scoring.
What I like: the docs say it needs nothing external, and all the data stays on your machine.
What I don't like: it is one more moving part, and I'd only add it once the built-in files feel too small.
Best for: the docs say it is best for local-only memory with advanced retrieval.
Cost: it is free.
How to get it: run hermes memory setup and choose holographic.
5. Honcho: 7.0 Out Of 10
Honcho is an external provider maintained by Plastic Labs that builds a model of the user across sessions.
What I like: each profile gets its own AI identity while sharing one view of the user, which suits a team of agents.
What I don't like: the cloud version sends your conversations to Honcho's servers and has its own pricing.
Best for: the docs say it is best for multi-agent systems with cross-session context.
Cost: it is paid on Honcho Cloud and free when you self-host.
How to get it: run hermes plugins install honcho and then hermes memory setup.
6. The Other Official Providers: 6.5 Out Of 10
The docs name seven more providers, and each has a clear niche.
| Provider | What the docs say it is best for | Storage | Cost in the docs |
|---|---|---|---|
| Mem0 | It suits hands-off memory with automatic extraction. | It uses the cloud, your own server or an in-process mode. | It is free or paid. |
| Hindsight | It suits knowledge-graph recall with entity relationships. | It uses the cloud or a local database. | It is free or paid. |
| Supermemory | It suits semantic recall with user profiling. | It uses the cloud or a self-hosted server. | It is free or paid. |
| OpenViking | It suits self-hosted knowledge management. | It is self-hosted. | It is free. |
| ByteRover | It suits developers who want local-first memory with a CLI. | It is local with optional cloud sync. | It is free or paid. |
| RetainDB | It suits teams already using RetainDB. | It uses the cloud. | It is $20 a month. |
| Memori | It suits agent-controlled recall with structured attribution. | It uses the cloud. | It is free or paid. |
What I like: there is a provider for almost every taste, and most have a free mode.
What I don't like: only one can be active at a time, and each adds setup work.
How to get one: run hermes memory setup, and install the plugin first if the provider lives in the catalog.
Which ones are bundled depends on your version, so run hermes memory --help to see your own list.
🔥 Want the exact memory setup that scored 9.0 here? Inside the AI Profit Boardroom, I've got step-by-step videos on connecting Obsidian to Hermes, plus the installable Agent OS with shared memory wired in. You also get four coaching calls a week with 3,400+ members. → Get access here
Hermes Memory Plugins: The Useful Extras
The plugin catalog is a separate thing from the providers above.
I ran hermes plugins search memory on 10 October 2026, and it listed 58 entries in the memory category.
I have not tested most of them, so I'm only naming the helpers whose catalog descriptions solve a clear problem.
| Plugin | Tier | What the catalog says it does |
|---|---|---|
| hermes-memory-wiki | Official | It adds a Memory Wiki dashboard tab and a read-only audit panel for MEMORY.md and USER.md. |
| memory-review | Community | It lets you tick staged memory writes and approve or reject them in a dialog. |
| memory-rewind | Community | It keeps an automatic version history of built-in memory so you can diff and restore. |
| memory-shield | Community | It keeps the agent from rewriting or wiping what it remembers about you. |
| obsidian-memory | Community | It lets you browse, search and edit Obsidian vault notes from the Hermes dashboard. |
Community plugins are third-party code, so read the disclosure on each one before you install.
My post on the Hermes Agent plugins marketplace explains the tiers.
The Commands You Need For Hermes Agent Memory
| Command | What it does |
|---|---|
hermes memory setup |
It opens an interactive picker for an external provider. |
hermes memory status |
It shows which memory provider is active. |
hermes memory off |
It disables the external provider and leaves built-in memory on. |
hermes memory reset |
It erases all built-in memory, so use it with great care. |
hermes journey |
It shows a timeline of learned skills and memories that you can edit or delete. |
/memory pending |
It lists saves that are waiting for approval. |
/new |
It starts a fresh session so new memory entries load. |
The Privacy Verdict On Hermes Agent Memory
Memory is a list of true facts about you, so it deserves more care than a chat log.
I found that out while filming my Herald OS video.
I opened the Memory page to show what Hermes remembered, and my Telegram details were visible on screen.
So my first rule is to check memory before you share your screen or record anything.
My second rule is to keep passwords, API keys and tokens out of every memory store.
My third rule is to read a cloud provider's data terms before you switch it on.
The docs say Hermes scrubs recognisable secrets before it hands data to a provider, but a scrub is a backstop and not a plan.
Local options score ten for privacy in my table for this exact reason.
You can see the rest of that interface in my Herald OS review.
What To Store And What To Skip
| Store it | Skip it |
|---|---|
| Store preferences, such as tone and format. | Skip passwords, keys and tokens. |
| Store environment facts and tool quirks. | Skip facts a quick search can find again. |
| Store corrections and project conventions. | Skip logs, code dumps and data tables. |
| Store completed work with a date. | Skip temporary paths and one-off debugging details. |
The docs recommend tidying up once a file passes 80% of its limit.
Anything long should go in the vault instead.
🔥 Not sure which memory option fits your setup? Inside the AI Profit Boardroom, you can bring your Hermes setup to one of four coaching calls a week and get a straight answer. There are 3,400+ members running these same tools. → Join the AI Profit Boardroom
Who Should Pick Which Hermes Agent Memory
Pick the built-in memory if you are new and run one agent.
Pick the vault if you run several agents or want your knowledge to outlive any one model.
Pick profiles if you serve separate clients.
Pick Holographic if you want deeper recall with no cloud.
Pick Honcho if you run a team of agents that all need to know the same user.
Pick one of the other providers only if its niche matches a real problem you have today.
Related Reading
- The best Hermes Agent setup ranks the ways to install and run Hermes.
- The best Hermes Agent skills covers the skills side of how Hermes learns.
- The Hermes Agent OS free options shows interfaces with a memory page built in.
Also On Our Network
- 🌐 The operator's step-by-step Hermes memory setup
- 🌐 Hermes memory for agencies, coaches and creators
- 🌐 The memory layer of an Agent OS, explained
- 🌐 Which models handle memory best, with GoldieBench scores
FAQ: Best Hermes Agent Memory
What is the best Hermes Agent memory?
My verdict is the built-in memory paired with an Obsidian vault.
It is free and private, and any agent can read the same notes.
How big is Hermes Agent's built-in memory?
MEMORY.md is limited to 2,200 characters and USER.md to 1,375 characters.
That comes to roughly 1,300 tokens in every session.
Which external memory providers does Hermes Agent support?
The docs name Honcho, OpenViking, Mem0, Hindsight, Holographic, RetainDB, ByteRover, Supermemory and Memori.
Only one of them can be active at a time.
Is Honcho or Mem0 better for Hermes Agent?
The docs say Honcho is best for multi-agent systems and Mem0 is best for hands-off memory.
I haven't run a long head-to-head test, so choose by which description fits you.
Is the built-in Hermes memory enough?
It is enough for one agent on one machine.
Add a vault or a provider once you want shared knowledge or more than a few paragraphs of facts.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built Goldie Agency into a seven-figure SEO and link building agency.
- I've grown my YouTube channel to 400,000+ subscribers.
- I run GoldieBench, where I score AI models on real one-shot builds.
- I wrote "Link Building Mastery", which is available on Amazon.
→ Get my best AI training inside the AI Profit Boardroom
My verdict is simple: keep the built-in files, add a vault, protect your privacy, and you have the best Hermes Agent memory.
