Hermes Agent is an open-source AI agent built over 9 months by the Nous Research team, and it has gathered 66,000 stars on GitHub. Its defining trait is "self-evolution": after completing a task, it automatically distills reusable Skills and keeps improving with subsequent use. It now natively supports WeChat — just scan a QR code to connect. This article walks you through installation and configuration from scratch.

What Is Hermes Agent

Simply put, Hermes Agent is an open-source, self-hostable AI agent. It can run long term on a local machine or in the cloud, and it connects to everyday chat tools like Telegram, Slack, Discord, WhatsApp, iMessage, Feishu, DingTalk, and WeChat. You talk to it directly in any chat window and let it complete tasks independently in the background.

Its differentiating core capability is the learning loop: after finishing a complex task, it automatically distills reusable Skills from it and saves them as standalone documents. In later use, these Skills load on demand and keep improving based on new usage feedback.

Combined with persistent cross-session memory, scheduled tasks defined in natural language, and a mechanism for running multiple sub-agents in parallel, Hermes Agent can run for the long haul and evolve continuously.

What You Need Before Starting

• OS: Linux, macOS, or Windows (Android phones via Termux also supported)
• Model API: at least one model API key with a 64K context window. GPT, Claude, GLM, MiniMax, Kimi, Qwen, DeepSeek, and others all work
• Cost: $3.99 to get started. A $5 server rental gets you 7x24 operation
• GitHub repo: https://github.com/nousresearch/hermes-agent

Step 1: Install Hermes Agent

Open a terminal and paste one command:

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

No need to worry about environment issues — whatever is missing, Python, Node.js, and the like, Hermes Agent installs it for you.

Tip: If you also run OpenClaw, Hermes detects it automatically and asks whether to import OpenClaw's settings, memories, skills, and API configuration.

Step 2: Configure the Model API

Once installation finishes, Hermes guides you through model configuration. Have the API key for the model you want to use ready.

Supported model providers include OpenAI, Anthropic, Groq, DeepSeek, MiniMax, GLM, Kimi, Qwen, and more — 13 in total. You can also connect a local model via the --ollama flag, no internet needed.

The model's context window must be at least 64K to maintain enough working memory.

To switch models, use:

hermes model

Step 3: Connect WeChat

This is currently the most anticipated feature. Hermes connects to WeChat through Tencent's official iLink Bot API — not a third-party reverse-engineered protocol.

Install Dependencies

Two packages are hard requirements:

pip install aiohttp cryptography

If you want the QR code displayed right in the terminal, add one more:

pip install qrcode

Log In by Scanning the QR Code

One command launches the setup wizard:

hermes gateway setup

Choose Weixin. The wizard automatically brings up a QR code; scan and confirm with WeChat on your phone. The account credentials are written to the ~/.hermes/weixin/accounts/ directory.

After successful confirmation, the terminal shows:

WeChat connected successfully, account_id=your-account-id

Configure Environment Variables

Open ~/.hermes/.env and at minimum fill in the account_id:

WEIXIN_ACCOUNT_ID=your-account-id

To restrict messaging to the bot to only yourself:

WEIXIN_DM_POLICY=allowlist
WEIXIN_ALLOWED_USERS=user_id_1,user_id_2

Group messages are off by default; enable the allowlist manually:

WEIXIN_GROUP_POLICY=allowlist
WEIXIN_GROUP_ALLOWED_USERS=group_id_1

Start the Service

hermes gateway

Send the bot any message from WeChat on your phone and you'll see a reply within seconds — even the "typing" indicator shows up in the chat.

Note: Try it with a secondary account first before hooking up your main one. A single WeChat message is capped at 4000 tokens; anything longer is split automatically, and long replies aren't a great experience right now.

Step 4: Daily Use

Start a Hermes conversation:

hermes

Once you see the welcome screen you can start chatting. When idle, Hermes rests automatically and consumes almost no tokens until you wake it again.

If you want to update to the latest version and enable the web management UI:

hermes update
hermes dashboard

Open 127.0.0.1:9119 in a browser to manage system status, browse past sessions, analyze token usage, and manage scheduled tasks and skill toggles. API key configuration finally has a visual interface too — no more hand-editing YAML files.

FAQ

• WeChat messages get split into multiple segments: A single WeChat message is capped at 4000 tokens and gets split automatically beyond that; there's no better solution yet.
• Disconnection (error code -14): The most common cause is an expired session. Just run hermes gateway setup again and scan a fresh code.
• Error "Another local Hermes gateway is already using this Weixin token": One token can only serve one poller; stop the other one first.
• Media file send/receive failures: First confirm cryptography is installed. WeChat's CDN uses AES-128-ECB encryption — without that library you can't even pull down images.

Core Features at a Glance

Official Resources

• GitHub repo: https://github.com/nousresearch/hermes-agent
• Official docs: https://hermes-agent.nousresearch.com/docs/getting-started/quickstart
• WeChat integration docs: https://hermes-agent.nousresearch.com/docs/user-guide/messaging/weixin