Hermes-Agent-Phone

What Hermes-Agent-Phone does

The app runs on Android 9 or later and exposes a REST API on the device's local network address, for example http://192.168.1.100:9527. A caller sends a natural-language prompt to /api/agent/execute_task, which waits up to 120 seconds for the result, and can check /api/agent/status or stop a task at /api/agent/cancel_task. Requests carry an X-APKCLAW-TOKEN header. The app can capture screenshots, parse the UI tree and perform click, swipe, input and long-press gestures. The LLM is configured inside the app with an API key, base URL and model name.

Its distinguishing feature, introduced in v6, is workflow templating: a successful task is saved as a template, and a similar later task can run from that template without the LLM planning again. The README says this saves time and tokens and documents template endpoints under /api/workflow. It also lists messaging channels (DingTalk, Feishu, QQ, Discord, Telegram, WeChat) and integration with the Hermes Agent CLI. The README is written mostly in Chinese.

Key features

• Natural-language task execution through /api/agent/execute_task
• Token-authenticated REST API over the local network
• Screenshots and UI tree parsing so the agent can read the screen
• Click, swipe, text input and long-press gestures
• Workflow templates saved automatically from successful runs
• Floating window that shows task status

When to use it

• Sending a message in a chat app by describing it in plain language
• Searching a shopping app and opening a product page
• Changing a device setting such as screen brightness
• Repeating a routine task from a saved workflow template

Who it is for: Developers who want a Hermes Agent or another AI agent to operate an Android phone they own through an HTTP API.

How it fits with Hermes Agent

The README presents it as an integration with the Hermes Agent CLI, where the API token configured in the app is used by Hermes to authenticate its calls to the phone.

Requirements: Android 9+, the Accessibility Service permission, an LLM API key, base URL and model name, an API token, and LAN Config switched on in the app

Note: The README warns that the app can control an Android device, so use it only on devices you own or are authorized to use and do not expose its LAN API to the public internet.

FAQ