Hermes: The Self-Improving Agent That Gets Smarter Every Day

追加: 2026-05-05

[00:00:00]Okay, Hermes is an open-source AI agent created by American company News Research that is self-improving. So, basically, the more you use it, the better it gets. It reflects, learns, and evolves on its own. It never forgets anything you said, and it even creates its own skills. But, is all of that enough to replace something like Open Claw, which supports many more channels, has better sandboxing, and is much more mature? Hit subscribe, and let's get into it.

[00:00:31]So, the name Hermes, surprise, surprise, comes from the Greek messenger god, and that's also where this symbol comes from. You'll see more of it later on in the video. But, as it stands, I've already made a video about Open Claw, which is great, but has a lot of features that I won't personally use, and Nano Claw, which has a much smaller feature set, but is built on top of the Clawed Agent SDK, which is now less usable for me because of the weird rules around using the Claw description with third-party tools. So, now I'm on the lookout for a new AI assistant, and let's see if Hermes, the self-improving AI agent, can fill that void. I'm going to use it to create promotional tweets for me based on past videos that I've created, and I'm going to give it some scripts and directions to get to that stage. Now, this is quite a small task, but the focus is more to see if Hermes can remember my writing style and all the feedback I'm going to give it to create a tweet that I like without me asking it over and over again. Let's go.

[00:01:29]So, I've already gone ahead and installed Hermes using this command, which is very simple and went through everything from collecting a model. I chose Open Research with Gemma 4, but if my hardware could handle it, I would run it locally and connect it to Hermes.

[00:01:42]Messaging platforms and tools for the CLI. If you've used Open Claw, this whole process will feel very familiar.

[00:01:48]I've also set it up on a VPS to be on the safe side, but if you wanted to, you could easily install it locally on your machine. So, from here, I'm going to write the Hermes command, which will start a new chat, showing the Hermes symbol with the available tools and skills over here. Note, when you run the Hermes command, it creates a new session and doesn't resume the previous one unless you specify, just like Claw Code.

[00:02:07]So, here, I'm going to give it a prompt.

[00:02:09]I want you to help me write tweets based on the scripts from my videos. Let's go through the process of doing that. After a while, it comes back with a response, which I like the structure of, and so I'm going to give it a follow-up prompt.

[00:02:19]I have scripts inside the scripts folder. Study them to understand my writing style and voice. I've also given it my target audience and the length I'd like my tweets to be. So, now it's using some tools to search through my files, and after a while, it analyzes my script to give me a breakdown of my style. So, it says I'm pragmatic and skeptical, which is true. I'm developer-centric, and I'm transparent and relatable. It's also come up with a strategy for my target audience, which I like the look of. But, I've changed my mind. Even though I did say I wanted the tweets to be around 210 characters, I actually want them to be a bit longer. So, I'm going to give it a new prompt, and I have noticed it's been taking a while and using a lot of context. So, what I can do is change the model mid-session by running the model {slash} command and specifying the model I want. In this case, I want GLM 5 Turbo. So, now it's switched to that model, I'm going to give it a new prompt to make the tweets longer, and it comes back with a response much faster, but has also added a lot of information to memory without me telling it. So, it's changed the length from 210 to 400, and has changed the style of tweets that I want. Let's see if it can actually generate a decent tweet from my latest script. And it has come up with a pretty decent first attempt, but there are a few things that I won't personally say, like breaking a sweat. And I wouldn't use the word incredible, I'd use the phrase really good. And after a few tweets, it's come up with a tweet that I would say I'd actually use in my profile, and it's saved that all to memory. I'm going to prompt it to create a skill, so it's easier for me to write tweets in the future. And now it sees the skill manager skill to go ahead and create a skill.

[00:03:50]Let's see this in action. And look at that, it's written a tweet for me with multiple options, and I can select the one I like the most. It's even gone ahead and created a thread that I can use to write multiple tweets if I wanted to. So, technically, because it's remembered everything, if I create a brand new Hermes session, change the model from the default, and ask it if it knows how I like to write my tweets, it comes back with a response telling me exactly how I like to write my tweets, even down to the type of emojis I like to use. Now, you may be wondering, how is Hermes able to pull all this information from memory without burning through your tokens? Well, memory is stored in an external file, so your memory.md file, or an external processor like Supermemory, Mem Zero, or Open Viking if you configure it. And memory is preloaded each session or pre-fetched, but it's not the full thing. In fact, it's a compacted version that's limited to about 3 and 1/2 thousand characters, which is roughly 700 tokens depending on the model. But, all sessions are stored inside an SQLite database using FTS5 for full-text search. So, if you ask Hermes to remember what you said yesterday, it will go into the database, do the search, and give you that information.

[00:05:02]It also does something a bit weird. It compresses your session above 50% context window, which is different from something like Claw Code, which does it at 80%. But, I guess it's difficult to tell a good measure depending on the model, so 50% is a good rough number.

[00:05:16]But, what it does is, instead of just compressing the whole thing, it removes the output from old tool calls and keeps the head and tail of the session, but compresses the middle. This is what actually gets saved in the SQLite database, not the full conversation itself. It also nudges itself every 10 or so turns to save important information to memory, and also to write a skill whenever that's necessary. Now, I know it's very difficult to see the full power of Hermes in this very short demo session that I gave it, but hopefully, you can kind of extrapolate how well it will remember and create skills based on the information you give it. And actually, I'm going to be using it more often. So, this month, or maybe the month afterwards, I'm going to focus on using Hermes as my main personal assistant with a very cheap model like GLM, and I'll let you know how it goes.

[00:06:02]But, as usual, let me know your thoughts in the comments. Again, don't forget to subscribe, and until next time, happy coding.