How to deploy a versatile Hermes that can do penetration testing and reverse engineering? One article tells you!

1. Introduction

If you’ve read my previous posts, you know I’ve been tinkering with Hermes lately.

It can now help me with many tasks, especially in reverse engineering, security analysis, and authorization testing. The experience is quite interesting.

My main model is deepseek-v4-flash, and its biggest advantage is high cost-performance and decent speed. However, DeepSeek itself has some shortcomings, such as not being suitable for direct image recognition, voice input, and voice output tasks.

So this article not only teaches you how to deploy Hermes, but also covers how I filled in these missing capabilities.

The final effect is roughly:

• Hermes runs on a cloud server, always online;
• Main model uses deepseek-v4-flash;
• Direct interaction with the Agent via Telegram Bot;
• Image recognition, voice parsing, and voice output are supplemented by other models;
• Finally integrated into reverse engineering and security analysis workflow.

2. Prepare a Cloud Server

First, you need to prepare a cloud server.

The configuration doesn’t need to be too extravagant; an ordinary VPS is sufficient. Because Hermes itself is mainly an Agent framework, the real cost usually comes from the large model API you call, not the server itself.

I usually use Oracle servers, but for this demo, I’m using an AWS Hong Kong server. The server OS is Debian.

Recommended configuration:

• CPU: 1 core or more
• Memory: 1GB or more
• OS: Debian 11 / Debian 12
• Disk: 10GB or more
• Network: Must be able to access GitHub and official installation sources normally

3. Connect to the Server

After getting the server, the provider usually gives you the following information:

• Server IP
• Username
• Password or SSH key

The first step is to connect to the server.

For SSH tools, I personally recommend Termius – it has a cleaner interface and is convenient for managing multiple servers.

Termius official site: https://termius.com/

After downloading and installing, open Termius and log in.

[Termius main interface]

Click New Host in the top-left corner to go to the add server page.

[Click New Host]

Fill in the information provided by your server provider:

• IP address
• Username
• Port number
• Password

The default SSH port is usually: 22

If you are logging in with a password, fill in the details and click Connect directly.

[Fill in server info. How to fill for key login?]

If your server uses key login, you need to import the key into Termius first.

Click Keychain on the left, then click New Key.

[Enter Keychain]

Here you need to fill in the key name and the private key content.

The public key is optional, but the private key is required. If you already have a key file locally, you can directly click:

Import from key file

[Import SSH key]

After importing, go back to the server configuration page.

In the password field, select: + SSH ID, Key

Then select the key you just imported.

[Select SSH Key]

Finally, click Connect to connect to the server.

4. Update the Debian System Environment

After successfully entering the server, update the system environment first.

[Successfully connected to server]

If you are not currently root, you can elevate privileges first:

sudo su

Then update the system package sources and base packages:

apt update && apt upgrade -y

[Execute sudo and system update]

Just wait normally for the update process.

Then install common dependencies:

apt install -y curl wget git vim nano unzip tar tmux htop build-essential ca-certificates sudo

These dependencies will be used later:

• curl / wget – pull installation scripts
• git – clone projects
• vim / nano – edit configuration files
• tmux – keep remote sessions from disconnecting
• htop – view server resource usage
• build-essential – compile some dependencies
• ca-certificates – handle HTTPS certificates
• sudo – privilege escalation for normal users

After the update is complete, it’s recommended not to keep using root user. You can switch back to a normal user.

For example, my normal user is admin:

su admin

Note: Replace admin with your own username.

5. Install Hermes

After the environment is ready, you can start installing Hermes.

The official provides a one-click installation command. Just enter:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

This command downloads the installation script from the official Hermes address and executes it with bash.

After installation is complete, the terminal will enter the first-time configuration page.

If it doesn’t start automatically, you can manually execute:

hermes setup

If you get a “hermes not found” error, refresh the environment variables first:

source ~/.bashrc

Then execute:

hermes setup

6. First-Time Configuration of Hermes

After entering the first-time configuration page, you’ll see several options:

Quick Setup
Full setup
Blank State

It’s recommended to choose the second one: Full setup

Because we will configure models, tools, and various capabilities ourselves later. If you only choose Quick Setup, it’s convenient but less controllable.

After selection, you’ll enter the full configuration process.

[Hermes configuration page]

Next, you’ll be asked to choose an API Provider. I’ll use opencode as an example.

After entering the provider selection page, choose OpenCode Go:

Then enter your opencode API Key.

[Enter API Key]

Press Enter to continue after entering.

For Base URL, you can press Enter to use the default configuration.

For model selection, I chose: deepseek-v4-flash

For Backend, choose: local

The subsequent options can be checked according to your needs.

If you just want to get Hermes running first, you can follow the default configuration and adjust later.

[Check options as needed]

For configuration items you’re unsure about, you can skip them for now.

After configuration is complete, you’ll see a completion prompt like below.

[Hermes configuration complete]

At this point, the basic environment of Hermes is set up.

You can check with the following commands:

hermes --help
hermes doctor
hermes

If these commands execute normally, Hermes has been installed successfully.

7. Configure Telegram Bot

After Hermes is installed, the next step is to configure the Telegram Bot.

This way, you can interact with Hermes directly on your phone without needing to SSH into the server each time.

Of course, besides Telegram, you can also choose QQ Bot, WeChat Bot, etc. Here I’ll use Telegram as an example.

Run:

hermes gateway setup

Enter the gateway configuration page.

[Gateway configuration]

Select Telegram, then you’ll enter the Telegram configuration process.

Here, choose to create the Bot using the first method, and the terminal will give you a link.

Copy the link and open it in your browser.

If your computer isn’t logged into Telegram, you can also open it on your phone browser.

After successful creation, you’ll see the Hermes Agent bot ready in Telegram.

[Telegram Bot created successfully]

Go back to the SSH page, and enter y as prompted.

Then enter your Telegram user ID.

You can get your user ID via the @userinfobot in Telegram.

Send it any message, and it will return your user ID.

[Get Telegram user ID]

Fill in the user ID, then continue by pressing Enter to confirm.

Finally, select done. Telegram Bot configuration is complete.

[Telegram allowlist configuration]

After configuration, start or restart the gateway.

8. Supplement Image Recognition and Voice Capabilities

Since my main model is deepseek-v4-flash, which has high cost-performance but does not support voice and image recognition.

So I will additionally connect other models to handle image recognition, audio parsing, and voice output.

Later, you can input corresponding prompts in the Telegram Bot so that Hermes automatically calls different models for different tasks.

1. Integrate Image Recognition and Audio Parsing

Prompt:

For all subsequent image recognition and audio/voice parsing tasks, please uniformly call the mimo-v2.5-flash model built into opencode.

2. Configure Image Recognition and Audio Parsing

Configure Voice Output Prompt (for free Xiaomi TTS, see my post: https://x.com/i/status/2072554282214432866):

All voice output in this session should uniformly call the interface address:
Use model: mimo-v2.5-tts-voicedesign
API Key: <Your API Key>
Design voice tone autonomously on first generation: 20-year-old young female voice, clear and natural. After the first voice tone is determined, use that voice line throughout this conversation without change.
All subsequent voice messages should be sent in bubble format.

9. Integrate Reverse Engineering Workflow

Prompt:

Clone/download the GitHub project repository:
https://github.com/zhaoxuya520/reverse-skill
Read the entire readme_ai.md file in the project thoroughly, without missing any instructions, configuration, or deployment steps.
Follow all the specifications and operational requirements noted in readme_ai.md strictly, and execute the associated processes in order.

Completion screenshot:

[Figure]

10. Agent Wallet and Autonomous Payment Capability

If you want to tinker further, you can also explore Agent wallet, autonomous payment, and autonomous use of paid services.

This kind of gameplay will be very interesting in the future, especially when the Agent itself calls services, purchases APIs, and completes tasks.

If you want the Agent to have a wallet, pay autonomously, transfer autonomously, etc., you can input the following prompt:

Run: https://agentpassport.ai/install.sh
Configure as instructed. Since my main agent is already configured, and the configuration process is very simple, I won't repeat it here. The operator is KiteAi (https://gokite.ai), with PayPal leading a $33M funding round – very safe and reliable!
In the future, agent payment will definitely become mainstream (not an ad, I genuinely think it's very useful!).
Payment arrival is extremely fast! And it enables the Agent to support many features, such as:

• Services payable directly with USDC (x402 protocol):

• Via Tempo protocol (requires cross-chain routing):

…and many more. You can explore yourself. I find it quite interesting, haha.

11. Finally, Test Hermes’ Penetration Effect:

For daily use, you can directly throw suspicious links (like scam URLs) to Hermes for analysis, reducing the chance of being scammed! Especially for unknown links involving on-chain transactions.

#Reverse #Penetration #Workflow #Hermes #DeepSeek