hermes-agent

中文版 | English

English

Sandbox runtime for AI agents driven by NousResearch Hermes models, supporting tool calling and multimodal tasks.

Environment

This template comes preinstalled with the following tools and dependencies:

Installation

Install from a local path:

abx template install ./hermes-agent --registry-type local

Install from GitHub:

abx template install alibaba-agent-sandbox/awesome-templates//hermes-agent

Usage

Create a sandbox instance and pass in the LLM configuration:

abx create --template hermes-agent \
    --env OPENAI_API_KEY=sk-xxx \
    --env OPENAI_BASE_URL=https://api.your-provider.com/v1

Quickly install agent frameworks with uv:

abx exec <sandbox-id> -- uv pip install langchain langgraph

Call a Hermes model via the OpenAI-compatible interface:

abx exec <sandbox-id> -- python3 -c "
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
    model='NousResearch/Hermes-3-Llama-3.1-8B',
    messages=[{'role': 'user', 'content': 'Write a file search tool'}],
    tools=[{
        'type': 'function',
        'function': {
            'name': 'search_files',
            'description': 'Search files',
            'parameters': {'type': 'object', 'properties': {'query': {'type': 'string'}}}
        }
    }]
)
print(response.choices[0].message)
"

Using the Python SDK:

from agent_sandbox import Sandbox

sandbox = Sandbox.create_sync(
    template="hermes-agent",
    env={
        "OPENAI_API_KEY": "sk-xxx",
        "OPENAI_BASE_URL": "https://api.your-provider.com/v1"
    }
)

result = sandbox.commands.run("python3 --version && uv --version")
print(result.stdout)

Configuration

Environment Variables

Managing Dependencies with uv

This template ships with uv, a high-performance Python package manager. Using uv instead of pip is recommended:

# Install packages with uv (much faster than pip)
abx exec <sandbox-id> -- uv pip install <package-name>

# Create a project with uv
abx exec <sandbox-id> -- uv init my-agent-project

Notes

• API key security: Never hardcode API keys into the template; always inject them via the --env option or environment variables.
• Model compatibility: Hermes models are accessed via the OpenAI-compatible interface; configure OPENAI_BASE_URL to point at the actual hosting address.
• Resource recommendations: 4096 MB of memory is sufficient for basic agent tasks; 8192 MB is recommended for browser automation or multimedia processing.
• ffmpeg support: ffmpeg is preinstalled for audio/video processing tasks.
• Virtual environment: Python packages are installed in /opt/venv; no manual activation is needed since PATH is preconfigured.

中文

NousResearch Hermes 系列模型驱动的 AI Agent 沙箱运行环境，支持工具调用（Tool Calling）和多模态任务。

环境说明

本模板预装以下工具和依赖：

安装方式

从本地安装：

abx template install ./hermes-agent --registry-type local

从 GitHub 安装：

abx template install alibaba-agent-sandbox/awesome-templates//hermes-agent

使用示例

创建沙箱实例并传入 LLM 配置：

abx create --template hermes-agent \
    --env OPENAI_API_KEY=sk-xxx \
    --env OPENAI_BASE_URL=https://api.your-provider.com/v1

使用 uv 快速安装 Agent 框架：

abx exec <sandbox-id> -- uv pip install langchain langgraph

通过 OpenAI 兼容接口调用 Hermes 模型：

abx exec <sandbox-id> -- python3 -c "
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
    model='NousResearch/Hermes-3-Llama-3.1-8B',
    messages=[{'role': 'user', 'content': '编写一个文件搜索工具'}],
    tools=[{
        'type': 'function',
        'function': {
            'name': 'search_files',
            'description': '搜索文件',
            'parameters': {'type': 'object', 'properties': {'query': {'type': 'string'}}}
        }
    }]
)
print(response.choices[0].message)
"

使用 Python SDK：

from agent_sandbox import Sandbox

sandbox = Sandbox.create_sync(
    template="hermes-agent",
    env={
        "OPENAI_API_KEY": "sk-xxx",
        "OPENAI_BASE_URL": "https://api.your-provider.com/v1"
    }
)

result = sandbox.commands.run("python3 --version && uv --version")
print(result.stdout)

配置说明

环境变量

使用 uv 管理依赖

本模板预装 uv 作为高性能 Python 包管理器，建议使用 uv 替代 pip：

# 使用 uv 安装包（速度远快于 pip）
abx exec <sandbox-id> -- uv pip install <package-name>

# 使用 uv 创建项目
abx exec <sandbox-id> -- uv init my-agent-project

注意事项

• API Key 安全：请勿将 API Key 硬编码到模板中，始终通过 --env 参数或环境变量注入。
• 模型兼容：Hermes 模型通过 OpenAI 兼容接口访问，需配置 OPENAI_BASE_URL 指向实际托管地址。
• 资源建议：基础 Agent 任务使用 4096 MB 内存即可；涉及浏览器自动化或多媒体处理时建议 8192 MB。
• ffmpeg 支持：已预装 ffmpeg，支持音视频处理任务。
• 虚拟环境：Python 包安装在 /opt/venv 中，使用时无需手动激活，PATH 已自动配置。

Deploy and local e2e

Install and register this template:

abx template install alibaba-agent-sandbox/awesome-templates//hermes-agent

From this directory (build, push, and register):

abx deploy

From the repository root, with agent-sandbox-server and httpx installed:

python3 scripts/run_local_e2e.py

That starts python3 commands.py in this directory and runs e2e_test.py.