In 2026, the pace of AI evolution is clearly noticeable.OpenClaw The crayfish craze hasn't faded yet, and now there's a new Hermes.
Some friends have already spent a night installing and debugging it, and report the experience is quite good~
Some friends say it's more fun than crayfish, and they've already started executing tasks.
They say it's stronger than OpenClaw because Hermes can evolve on its own~ I see!
Group members tested it: if for personal use, it can completely replace OpenClaw~ It seems,
in the AI era, a hit product is bound to emerge within two months.
But there are also friends whose installation got stuck and who found Doubao unhelpful.
Today, Sister Orange will walk everyone through understanding Hermes Agent, local installation and deployment, and the full Feishu integration process in one article.
Scan the QR code to follow the official account and reply with the keyword: "Agent”，Join the AI exchange group to get help with installation questions
What is Hermes Agent?
Hermes Agent is an AI agent project launched by the well-known open-source AI lab Nous Research .
GitHub address:https://github.com/NousResearch/hermes-agent
Its positioning is very clear: it is a persistent personal agent deployed on your local machine/server that grows together with you.
It was only officially open-sourced at the end of February this year, and in less than 2 months, its GitHub stars have already surged to 88k+, making it the fastest-growing AI Agent framework this year.
MIT license, completely free, commercially usable, supports Linux, macOS, and WSL2.
But what's truly impressive about Hermes Agent is actually these:
01 Write its ownSkill。
This is the core capability that sets it apart from other Agents.
After each completion of a complex task using more than 5 tool calls, it automatically distills the solution into a structured Skill document stored locally, so the next time a similar task comes up it can call it directly, skipping the exploration phase—the longer you use it, the faster it gets.
Official testing shows that after using it for a week, the execution speed of research-type tasks can increase by 40%.
It also regularly self-reflects, automatically optimizing the Skills it has already generated every 15 tasks, with no manual maintenance required at all.
It's like a student who seriously keeps an error notebook—next time they encounter a similar problem-prone question, they'll be more careful.
02 Three-layer memory。
This isn't simply memory that stores chat records. Hermes's memory is divided into three layers:
Skill memory: the accumulated Skill documents.
Long-term memory: user habits, decision preferences, and historical task records.
Short-term memory: the current conversation context.
For example, if you chat halfway on Telegram and then switch to Feishu, it fully remembers the context; if you change computers, logging into the account lets you continue the conversation.
03 Security protection。
A single Gateway processhandles all platform integrations. It currently stably supports Telegram, Discord, Slack, Feishu, and WeCom, with Beta support for WhatsApp and Signal, so domestic teams can use it with confidence too.
Security is handled very solidly. Out of the box, it comes with five layers of protection: prompt injection scanning, credential filtering, sandbox isolation, operation approval, and IP whitelisting, so it's safe to use in production environments as well.
Seeing this, don't you want to deploy this horse 🐎 and give it a try right away?
Local deployment of Hermes Agent
I'll use the Mac system as an example to teach you. Open the terminal command, directly enter the official installation command below, and press Enter:
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bashIt probably won't go very smoothly, if you encounter the same error message as I did
The reason for the error is that although I had the proxy turned on, the terminal wasn't going through the proxy, so the Mac couldn't connect to GitHub.
💡Solution: If you have a proxy, just copy the terminal command shown here from the proxy and paste it into the terminal, then press Enter.
Alternatively, you can run the following command in the terminal:
#假设你的代理是7890端口（7890要替换成你的代理正确的端口）export https_proxy=http://127.0.0.1:7890export http_proxy=http://127.0.0.1:7890
Then run the Hermes Agent installation command above again.
When the terminal displays the Hermes Agent style shown in the image below, it means the installation has begun.
I had installed OpenClaw before, so all the required environments were ok, and it went very smoothly.
Next, Hermes Agent will automatically install some dependency components. The installation script will configure these things. It's all automated, so you don't need to worry about it—just wait.
Once it finishes running, it will display the configuration wizard state, and that's it.
Now it prompts that it detected I had previously installed OpenClaw, and asks whether to import its settings, such as model settings, API keys, and so on. I'll choose not to import first, entering "n" and pressing Enter.
If you want to import later, you can also manually migrate using the hermes claw migrate command. Judge for yourself here.
Next, select "Quick Step"
Then you'll get to the model API configuration. Just select a model that already has an API key.
Then for the IM configuration, you can skip it for now. After Hermes is up and running, let the Agent help you configure the integration.
For the subsequent choices, you can just go with the defaults.
The prompt shows: chat with Hermes now? Enter "y"
Then you can chat with Hermes. Enter "Hello", and if you get a reply, it's ok.
Integrating Feishu
Log in to the Feishu Open Platform and create an enterprise self-built app.
https://open.feishu.cn/
Enable the "bot" capability, and apply for some message permissions such as receiving messages and sending messages. This is the same as the previous OpenClaw integration process.
After creating and publishing, give the Feishu bot's App ID and App Secret to Hermes Agent to help you configure the integration.
If it doesn't work in the middle, keep telling Hermes Agent what the current status is. For example, if sending a message to the bot gets no reply, tell it and let it solve it on its own.
In the end, once it's working, sending a message to Feishu will get a pairing code. Send that to Hermes Agent for pairing, and it's done.
Now I'm someone who has both the 🦞Lobster Agent and the Hermès Agent, hahaha.
Some takeaways
As a popular open-source AI Agent just like OpenClaw, the two have completely different positioning. Neither is better—it's just about which one suits your scenario better. If you have a lot of repetitive work to do every day, I'd recommend Hermes Agent more. Its built-in automatic learning ability means that the longer you use it, the better it understands your habits, and execution efficiency will improve noticeably. But if you have enterprise-level needs, or you're already using OpenClaw for multi-channel management, then continuing to use OpenClaw is fine too—there's no need to switch back and forth.
Hermes is more like a senior colleague. You don't need to explain things over and over, making it suitable for individuals and small teams that don't want to tinker and want to save time.
OpenClaw is like a controllable new employee. Its behavior is completely transparent and customizable, but it requires manual training and input, making it suitable for complex enterprise-level scenarios.
Overall, the deployment process is relatively simpler than OpenClaw's, though it might also be because I've already been down the Lobster path once before.
For about half an hour I ignored Hermes on Feishu, and then I received a message it sent proactively. In that instant, it really felt like a colleague, a friend, a partner, rather than a robot.
I want to share again a passage I wrote in an article before,that is this one。
Building an Agent is like raising a person, from child to adult。
I think it's very similar:
Ten months of pregnancy 🫄👉birth 🐣👉nurturing 🧑🍼👉learning to speak and express👉able to ask back and think👉solving problems on one's own👉learning skills Skills (walking, running, riding a bike, learning all kinds of life skills, social skills... etc.)👉the next step is to use skills Skill to find a job, earn money, get married, have children, and then survive.
Looking at the current direction of AI evolution, it matches completely.
Our generation isbridging the past and the future。
Scan the QR code to follow the official account, and reply with the keyword: "Agent”，Join the AI exchange group to get help with installation questions
Original link:Understand Hermes Agent in One Article, Local Deployment, Full Tutorial for Connecting to Feishu