Skip to main content

Hermes Agent

Configure and use in Hermes Agent

This tutorial is suitable for users who want to run Hermes directly on their computers.

Windows 用户

Windows recommends running Hermes in WSL. The following Windows command refers to the Windows + WSL environment by default; if you use the official Hermes Windows installation method, please press the official installer and refer to the same configuration file content after completing it.

Step 1: Install Hermes

Run the install command based on your system first.

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

After installation, Hermes 'configuration is usually located at:

~/.hermes

If thehermescommand cannot be found in the current terminal after installation, please close the terminal and reopen it.

Step 2: Connect Hermes to PuLu AI

First open the Hermes configuration directory. You can modify files with your handy editor.

Windows + WSL
cd ~/.hermes
explorer.exe .

打开 Hermes 配置目录

Then openconfig.yamland change the model configuration to the following:

model:
  provider: custom
  default: gpt-5.4-xhigh
  base_url: https://www.puluai.com/v1
  api_mode: chat_completions

It can be directly understood as follows:

  • base_urltells Hermes which interface address to send requests to in the future.
  • defaultis which model is used first by default.
  • api_modeJust write it first, without first studying the details of the protocol behind it.

Then open~/.hermes/.envand fill in your own key:

OPENAI_API_KEY=你的密钥

If there is alreadyOPENAI_API_KEYin this file, change it to the Key you want to use this time.

Step 3: Apply a cache compatibility patch

This step is used to make Hermes more suitable for interfaces such as PuLu AI and improve cache hit performance in repetitive context scenarios.

Windows + WSL
git clone https://github.com/foryourhealth111-pixel/hermes-codex-proxy-cache-compat.git
cd hermes-codex-proxy-cache-compat
bash scripts/apply_patches.sh ~/.hermes/hermes-agent
bash scripts/install_skill.sh ~/.hermes

Info

The transit address not only depends on the content sent, but also is affected by the complete request structure. The patch will fill in this structure, bringing cache hits closer to the performance of native Codex.

Step 4: Start Hermes and make a minimum available confirmation

Windows + WSL
hermes

After entering, ask a simple question first, such as:

Hermes 运行示例

请用三句话介绍一下 Hermes 是做什么的。

As long as it can return content normally, this round will be considered complete.

If this step reports a Key error, an address error, or the model does not exist, go back to the second step and check again.

Tips

If you want to check if the cache is active later, you can ask questions with similar prefixes twice in the same session, and then pay attention tocached_tokensor_PULUAI_TOKEN_1__.

Copyright © 2026-present PuLu AI