Hermes vs Claude Code

“Bro, you have got to try Hermes. It’s so easy to set up, and honestly, there’s nothing like it.”

“Okay… what can you do with it?”

“I text it from Telegram and it just runs. Morning briefing on the emails i got last night. Digs through my old notes when I ask. Writes its own little scripts when it gets stuck. It’s on all day.”

“Can’t the latest Claude Code do all of that?”

“But bro…”

“What?”

If you’re as lost as these two, don’t worry. Read on as we break down the key differences between Hermes and Claude Code.

The Quick Version

First, for the TL;DR crowd, here’s the whole comparison at a glance.

Hermes vs Claude Code — A Deep Dive

Claude Code and Hermes aren’t really the same category of product. One is a coding tool, the other is an always-on personal agent. But their use cases overlap quite a bit, and that’s where the comparison gets interesting.

What Each One Actually Is

Claude Code is Anthropic’s agentic coding tool. It has a desktop app and can also run in your terminal or IDE. It reads your codebase, edits files, runs commands, and handles git for you. It’s one of the best harnesses out there for writing software (if not the best).

Read more: Why Claude Code tops our tool rankings and the best Claude Code features you didn’t know exist.

Hermes is a different animal. It’s an open-source agent from Nous Research, released under the MIT license, that you run on your own machine. Instead of opening it for a task, you leave it running.

It talks to you over Telegram, Discord, Slack, WhatsApp, Signal, and a terminal. Its docs are blunt about the distinction: it’s not a coding copilot tied to an IDE, it’s an agent that “gets more capable the longer it runs.” It can write code, sure, but coding is one thing on a long list.

How the Memory Models Differ

Both tools remember things, and both lean on plain markdown files to do it. That’s where the similarity ends.

Claude Code primarily remembers the project you’re working on, while still carrying some preferences with you across projects. You can define project-specific conventions and instructions in a CLAUDE.md, which it loads whenever you work in that repository.

It also maintains an automatic MEMORY.md for each repo, where it records useful context it picks up along the way, such as your role, preferences, and corrections.

For preferences you want Claude Code to apply across repositories, you can use ~/.claude/CLAUDE.md and files in ~/.claude/rules/.

Hermes takes a different approach. Rather than centering its memory on a repository, it builds context around you. It curates its own MEMORY.md for notes and a separate USER.md that builds a profile of who you are and how you work.

On top of that, it saves every conversation from every surface into a local database with full-text search, so it can pull up something you said three weeks ago on Telegram whenever needed.

The real difference isn’t the file format. It’s the scope of the memory. Claude Code’s memory is largely organized around the projects you work on, while Hermes keeps accumulating context across your conversations and apps over time.

So if you want a tool that knows your project cold, Claude Code’s model fits. If you want an agent that gets to know you better the longer you use it, that’s Hermes.

You Can Run Almost Any Model on Hermes. On Claude Code, You Can’t.

This is one of the biggest differences between Claude Code and Hermes.

Claude Code runs Claude, and only Claude. You can route it through Amazon Bedrock or Google Vertex, but those are just different routes to the same Anthropic models. There is no official way to point it at GPT, Gemini, Llama, or a model running on your own GPU.

Hermes will talk to almost anything. Its rule is simple: if a server speaks the standard /v1/chat/completions format, Hermes can use it.

That covers 45-plus providers, including Anthropic, OpenAI, Google, DeepSeek, and xAI, plus local models through Ollama or vLLM. You switch models with one command and no code changes. The one real constraint is that the model needs at least a 64,000-token context window, or Hermes refuses to start.

Hermes Writes Its Own Skills, Unprompted

Hermes can write skills unprompted, by default. After it grinds through a hard task, it decides on its own to save a skill so it doesn’t have to figure the same thing out next time.

Claude Code can also create skills but only when you ask it to, or when you set up a hook to do it. It won’t automatically create one after completing a task. That automatic skill creation is a big part of Hermes’s “self-improving” approach. When it works well, Hermes becomes more useful the longer you use it.

When it doesn’t work, though, things can get messy. In our tests, Hermes sometimes created skills that referenced tools or environment variables that didn’t exist.

We also saw it get stuck repeating the same tool call over and over when there wasn’t a clear limit on how many times it could try. So while the self-improving loop is useful, it still needs some human oversight to make sure it’s learning the right things.

Always-On and Everywhere vs. Open-When-You-Need-It

Hermes is designed to run in the background. It stays active as a continuous process and has a built-in scheduler, so it can send you a morning report or run a nightly backup without you starting it each time.

Nous markets it as “cheap enough to leave on a five-dollar VPS”, hibernating when idle and reaching you through whatever chat app you already use.

Claude Code can also work beyond your desktop. Remote Control lets you control a session from your phone, Channels can send Telegram messages or webhooks into a running session, and cloud Routines can handle scheduled jobs.

But these features are still built around coding sessions, and don’t really make Claude Code an always-on personal assistant.

Hermes Makes Security Your Job

Hermes is open-source and self-hosted, so you have full control over how you run it, but that also means that you are responsible for securing it. If you give an agent access to your system and let it act without approval, you need to be careful about what it can access and what it is allowed to do.

Hermes does include security controls, such as approval settings and a command scanner. But you have to configure and manage them yourself.

Claude Code takes a different approach. It comes with Anthropic’s built-in safeguards and asks for permission before taking certain actions by default.

Can You Use Them Together?

Yes. Hermes is model-agnostic, so one of the models you can point it at is Claude. It can also use your terminal directly, which means it can run the claude command as one of its own tools.

In a setup where you have both, Hermes can be the always-on outer layer that reaches you on your phone and schedules the work, and Claude Code can be the heavy coding engine it calls when there’s actual software to write.

Why Would You Need Hermes if You Have Claude Code?

• You want an endlessly customizable personal assistant that’s on your phone and across all your apps, at all times.
• You want to run open-source or local models, or keep everything on your own hardware for privacy.
• You want something no vendor can switch off, and you’re willing to host and secure it yourself.
• You want an agent that builds up memory about you across every app you use, for as long as you run it.

Why Would You Need Claude Code if You Have Hermes?

• You write and ship software, and you want the strongest coding output with the least fiddling.
• You want deep IDE, terminal, and git integration built specifically for development.
• You want a permission model that’s safe by default and a vendor handling the backend.
• You want consistent results without having to choose or switch between models.

Wrapping Up

If you mostly write code and want great results quickly and consistently, Claude Code is enough on its own. If you want a persistent assistant that lives across your apps and runs your day, you should definitely install Hermes and play around with it.