What Is an Agent
This documentation keeps using words like "Agent," "drive," and "host." If you're new to this, it can feel abstract. This primer explains those concepts using real examples from the yoooclaw ecosystem.
What an Agent is
An Agent is an AI program that can autonomously call tools and complete multi-step tasks — not just chatting, but actually doing things for you: reading files, running commands, querying data, calling APIs, working step by step toward a goal without needing you to micromanage every step.
The difference from a plain chatbot:
| Plain chat | Agent | |
|---|---|---|
| Input/output | You ask, it answers | You give a goal, it breaks it into steps on its own |
| Can it act | Only gives advice | Calls tools to actually execute (read/write files, run commands, send requests) |
| Do you need to watch it | Every turn needs your input | Can run autonomously and continuously until done or stuck |
Which Agents show up in this documentation
- Claude Code / Codex: coding Agents that run in a terminal, able to read/write code and execute commands. The standalone CLI is designed for this kind of Agent — it packages notification, recording, and light-control capabilities as commands that an Agent can just call from its command list, no need to understand the internals.
- OpenClaw / QClaw: a class of "Agent host" apps — smart-assistant clients that can chat with you and get things done on their own. The OpenClaw plugin installs into this kind of host and adds phone notifications, recording transcription, and hardware light control on top.
- Hermes Agent: another Agent runtime. The Hermes plugin is the bridge between it and yoooclaw, and additionally ships an APP adapter that lets the phone app chat with Hermes directly.
A few high-frequency terms
- daemon: a program that runs resident in the background — you don't need to keep a terminal open and watch it. The standalone CLI ships its own local daemon, responsible for keeping the Relay connection alive and persisting notifications in the background.
- host: the container program an Agent actually runs inside, e.g. OpenClaw, QClaw, or Hermes Agent. A plugin runs inside the host, not as its own standalone process.
- MCP (Model Context Protocol): a standard protocol that lets an Agent call external tools/data sources uniformly, without writing custom adapter code for every single tool.
- Skill: an "instruction manual" (typically a
SKILL.mdplus some scripts) fed to an Agent, teaching it how to correctly call a specific command-line tool. The standalone CLI's Agent Skill is exactly this pattern. - Relay / tunnel: if the phone and the local daemon / host plugin aren't on the same network, you need a hosted WebSocket tunnel to forward requests in. All three forms (CLI, OpenClaw plugin, Hermes plugin) share the same underlying Relay implementation.
Next steps
- OpenClaw / Hermes Ecosystem Overview — how these names actually relate to each other.
- Getting Started — pick the standalone CLI, the OpenClaw plugin, or the Hermes plugin based on your situation.