Hermes Agent is an open-source AI agent framework developed by Nous Research, directly comparable to OpenClaw. It lets you set up an AI assistant on your own computer, VPS, or cloud environment—one that remembers, uses tools, and gradually gets better at understanding you over time.

It supports multi-platform interactions such as CLI, Telegram, Discord, Slack, WhatsApp, and more. You can connect model providers like OpenAI, Anthropic, OpenRouter, and Ollama, and it can also run terminal commands, install Skills, and set up scheduled automation tasks.

Even better, Hermes Agent’s official documentation provides a pipeline to connect OpenClaw memory and Skills to Hermes. This article will walk you through the complete installation and configuration process for Hermes Agent, as well as the tutorial for migrating OpenClaw memory and Skills into Hermes. Works with macOS, Linux, and Windows (WSL2).

What is Hermes Agent? How is it different from OpenClaw?

Most chatbots work like this: you ask a question, it answers, and once the interaction is done, that’s it. Hermes Agent is different. Its design is a “persistent agent system”: you can run it long-term on your own machine, VPS, or a remote environment, and talk to it via CLI, Telegram, Discord, Slack, WhatsApp, Signal, Email, or Home Assistant. Directly comparable to OpenClaw.

(AI also has to love Hermes! OpenClaw that evolves on its own: What is Hermes Agent? )

The core of Hermes Agent is long-term operation and building capabilities. The agent can not only execute tasks, but also convert experience into reusable skills, and continuously optimize its decision-making process through a memory system—gradually building an understanding of your behavior and preferences.

This “Closed Learning Loop” is the key difference with Hermes Agent. After completing tasks, the system automatically organizes the workflow, generates skill files, and then directly calls and optimizes them in later similar situations. Combined with its long-term memory mechanism based on SQLite and full-text search (FTS5), it has the ability to carry over across sessions, so it doesn’t have to start from zero every time.

Pre-installation prep

The installation environment in Hermes’s official Quickstart is straightforward: it supports Linux, macOS, and WSL2. Native Windows isn’t in the main recommended scope. If you’re a Windows user, it’s recommended that you install WSL2 first, then run the installation commands inside WSL2. After installation, you’ll also need to reload the shell, such as running source ~/.bashrc or source ~/.zshrc.

If this is your first time working with this kind of tool, you can initially think of Hermes as an “AI assistant that requires terminal operations.” You don’t need to understand every feature at first—just install it, choose a model provider, and then open the CLI and say its first message.

Step 1: Install Hermes Agent

First, open your terminal and enter the official one-line install command:

Linux / macOS / WSL2 curl -fsSL | bash

This is the first step of the official Quickstart. The overall design direction is similar to the “install first, then figure it out later” flow used by many open-source tools in recent years—first get the hermes command, then do the rest of the setup slowly.

Step 2: Configure the model provider

After Hermes is installed, the next step is to select a model provider using interactive commands. The three core commands the official provides are:

hermes model

hermes tools

hermes setup

hermes model is used to choose your LLM provider and model. hermes tools is used to configure which tools are enabled. hermes setup will configure the entire environment in one go.

Hermes supports many providers. If you want zero setup and the simplest path, you can choose Nous Portal. If you already have a ChatGPT ecosystem, you can try OpenAI Codex. If you’re a Claude user, you can directly choose Anthropic. If you want to go with an open multi-model route, OpenRouter is a convenient entry point.

One particularly notable feature is that Hermes also supports a Custom Endpoint. That means you can directly connect to Ollama, vLLM, SGLang, or any OpenAI-compatible API. This is very appealing if you want to run models locally or host your own inference service.

A very important advantage is: if you want to switch models later, you don’t need to change any code. Just run hermes model again.

Step 3: Start chatting

Once the model is set up, just enter:

hermes

You’ll enter Hermes’s CLI interface. The official documentation mentions that after startup you’ll see a welcome banner showing the currently used model, available tools, and installed skills. Then you can type messages directly—for example:

What can you help me with?

Hermes isn’t limited to text responses. According to the Quickstart, it will bring tools like web search, file operations, and terminal commands right into the mix. In other words, after you open it, you’re not just “chatting”—you’re already in an agent interface that can do real work.

Step 4: Try out Hermes’s core features

If you install Hermes and don’t know what to say first, the official has already provided some test-friendly examples. For instance, this one is great for verification:

What’s my disk usage? Show the top 5 largest directories.

Hermes will run terminal commands on your behalf, then organize the results for you. This is important because it shows that Hermes isn’t only good at answering knowledge questions—it can start touching your working environment.

The CLI also includes a full set of slash commands. After typing /, you’ll see an autocomplete menu. For example, /help shows all commands, /tools lists available tools, /model lets you interactively switch models, /save saves the conversation, and even /personality pirate provides a personality mode.

If you need to paste a longer prompt or code, you can use Alt+Enter or Ctrl+J to insert line breaks. If it takes too long, you can interrupt it by entering a new message or pressing Ctrl+C. After exiting, if you want to return to the previous session, you can use:

hermes –continue

or the short version

hermes -c

This is also very useful, because for many agent-like tools, the hardest part isn’t the first startup—it’s whether you can resume context when you come back the second time. Hermes does a very complete job in this area for the CLI experience.

Step 5: Connect to communication platforms like Telegram

A big feature of Hermes is that it’s not limited to the terminal. You can connect it to Telegram, Discord, Slack, WhatsApp, Signal, Email, or Home Assistant through a gateway—effectively bringing the same agent into the interfaces you already use every day. The gateway setup entry command from the official Quickstart is:

hermes gateway setup

This command will take you into the interactive platform configuration flow.

Step 6: Enable voice mode

If you want Hermes to not only interact with text, but also listen to your microphone and talk back, the official provides voice mode. First, install the voice-related packages:

pip install “hermes-agent[voice]”

If you want a free, local speech-to-text option, the official also recommends installing:

pip install faster-whisper

Then, after entering the Hermes CLI, type:

/voice on

After that, you can press Ctrl+B to record, or use /voice tts to have Hermes read the replies out loud. This isn’t only for the CLI—it can also extend to Telegram, Discord, and even Discord voice channels.

This kind of experience is actually very similar to upgrading an agent from a text-based chatbot to an AI version of Siri.

Step 7: Install Skills, schedule automation, and advanced play

Another strong point of Hermes is its Skills system. You can search for and install existing skills, such as:

hermes skills search kubernetes

hermes skills search react –source skills-sh

hermes skills search –source well-known

hermes skills install openai/skills/k8s

hermes skills install official/security/1password

hermes skills install skills-sh/vercel-labs/json-render/json-render-react –force

The official documentation reminds you that using –source skills-sh can search public skills.sh directories, while –source well-known can explore skills from the website’s /.well-known/skills/index.json. As for –force, you should only use it after you’ve reviewed a third-party skill.

In addition to Skills, Hermes can also schedule automation tasks. The Quickstart example even directly gives you a natural-language requirement:

Every morning at 9am, check Hacker News for AI news and send me a summary on Telegram.

Hermes will use the gateway to set up a timed cron job for you. This means it’s not just a reactive tool—it can proactively run tasks for you.

Also, if you care more about security isolation, you can move the terminal backend to Docker or SSH:

hermes config set terminal.backend docker

hermes config set terminal.backend ssh

Step 8: Use Hermes in an editor (ACP)

If you want Hermes not only to be a CLI agent, but also to enter your editor, the official supports ACP-compatible editors such as VS Code, Zed, and JetBrains. The installation method is:

pip install -e ‘.[acp]’

hermes acp

With this, Hermes can act as an ACP server and provide capabilities externally. For developers, this is like connecting the same agent “brain” to different working interfaces.

Step 9: Connect external tools with MCP

Hermes also supports MCP, which is the Model Context Protocol. You can treat external tools as plugins and plug them into Hermes. The official demo is adding a GitHub MCP server to ~/.hermes/config.yaml:

mcp_servers:

github:

command: npx

args: [“-y”, “@modelcontextprotocol/server-github”]

env:

GITHUB_PERSONAL_ACCESS_TOKEN: “ghp_xxx”

This means Hermes can not only use built-in tools, but also gain more external capabilities through the MCP ecosystem. For advanced users, this will be one of its most fundamental differences from ordinary chatbots.

Seamless migration tutorial: OpenClaw memory and Skills moved over to Hermes

Hermes’s official provides a complete OpenClaw migration mechanism that allows users to painlessly transfer existing agent settings and memory to a new system. Simply run hermes claw migrate in the terminal, and it will read data from the default ~/.openclaw/ directory and import everything into Hermes, including the personality (SOUL), long-term memory (MEMORY, USER), skills, model and provider settings, communication platforms (such as Telegram, Discord), and related API keys (optional).

If you want to preview the migration contents first, you can use the –dry-run mode to check changes, or use –preset full for a complete migration (including keys).

During the migration process, Hermes will re-parse and integrate OpenClaw’s memory data—for example, multiple memory files will be merged, deduplicated, and then written into the new memory system. Skills will be unified into Hermes’s skills directory, and conflicts will be handled according to your settings (skip, overwrite, or rename).

Model settings, inference parameters, session reset rules, and even MCP server and TTS configurations will be converted to Hermes’s config structure. At the same time, any partially incompatible items from the old system (such as plugins, hooks, or complex channel configurations) will be stored in an archive for later manual adjustments.

After migration is complete, the official recommends checking the migration report, verifying that the API key is working properly, restarting the gateway, and testing the communication functionality to ensure the whole agent runs correctly in the Hermes environment. Overall, this migration mechanism isn’t just an “integration”—it fully converts and rebuilds OpenClaw’s capabilities, memory, and settings into Hermes’s architecture.

Hermes Agent Frequently Asked Questions (FAQ)

Q: I’m a Windows user. Can I run the curl install command directly?

A: It’s not recommended to run it directly in CMD or PowerShell. First install WSL2 (Windows Subsystem for Linux). Hermes needs a Linux environment to fully support terminal tools and script execution. After installing WSL2, enter your Ubuntu terminal and then run the install command.

Q: After running the install command, I get “command not found.” What should I do?

A: Remember to run source ~/.bashrc (or source ~/.zshrc if you’re using zsh) to reload the PATH environment variables. 2. Models and Providers (Models & Providers)

Q: If I want to switch to a model from OpenAI or DeepSeek, what should I do?

A: Just type hermes model anytime. This will open an interactive menu, letting you reselect a provider or input a new API Key. Quick tip: If you have your own self-hosted local model (like Ollama), you can choose Custom Endpoint and enter your local Base URL.

Q: Does Hermes support “free” models?

A: Yes. You can access a variety of open-source—and possibly free—models via Hugging Face or OpenRouter, or you can use a Custom Endpoint to connect to an LLM you run locally.

Q: Do I have to start over every time I exit? How do I continue my previous conversation?

A: No. Use hermes –continue or the shorter hermes -c to jump straight back to where you left off in the previous conversation.

Q: Will Hermes really “mess up” my computer? Is running terminal commands safe?

A: Having AI run commands directly does involve risk. Recommendation: If you’re not sure what the AI might do, you can set up a sandbox environment (Sandboxing): hermes config set terminal.backend docker. This way, all its commands run inside an isolated Docker container and won’t touch your host machine files.

Q: Can I make Hermes reply to me on Telegram or Slack?

A: Yes! Enter hermes gateway setup, and it will guide you through connecting the messaging platform. After it’s configured, you can even give it instructions from your phone.

Q: What are “Skills”? How do I install them?

A: Skills are built-in functional modules (such as Kubernetes management and React code generation). You can use the following commands: Search: hermes skills search [keyword]. Install: hermes skills install [author/name]. View in the conversation: just enter /skills or /tools in the chat.

Q: How can I make it “run tasks on a schedule”?

A: Just tell it in plain language, for example: “Every morning at 9, check Hacker News for AI news and send it to me on Telegram.” Hermes will automatically write the Cron Job for you and set it up to trigger automatically.

Quick reference: Common commands

Command Function description hermes Starts the Agent Start a conversation /model Switch models instantly during the conversation /save Save the current conversation log hermes doctor (Emergency aid) When the program gets weird, use it to diagnose the problem hermes update Upgrade to the latest version Alt + Enter Enter multi-line text (suitable for pasting code)

This article: Complete tutorial—Build a growing AI assistant with Hermes Agent; Seamless migration of OpenClaw memory; First appeared on LianXin ABMedia.