Build Your First AI Agent

Key ideas

Agent loop

The agent reads your request, plans, uses a tool, checks the result, and repeats until the job is done.

Tools

The agent's hands: run commands, read files, search the web, send messages.

Skills

Saved recipes that teach the agent your way of doing a repeated job.

Memory

Notes the agent keeps between sessions so it remembers your projects and preferences.

Channels

Where you talk to it: the terminal, Telegram, WhatsApp, and other chat apps.

Scheduled jobs

Recurring tasks the agent runs on a timer, like a morning summary or a weekly report.

A practical starting sequence

1. Choose one tool from the Starting Guide — Hermes Agent for operational work or OpenClaw for a chat-first personal assistant are good defaults.
2. Install only from the official source linked there, and read its installation page end to end first.
3. Connect one model provider (an API key or sign-in) and run your first chat inside a disposable test folder.
4. Enable the minimum tool set — usually files and terminal — and keep approval prompts on.
5. Give the agent one clear job with a definition of done, and watch how it plans and uses tools.
6. Save one repeated workflow as a skill, then check that the agent reuses it next time.
7. Connect one chat channel such as Telegram, with sender allow-listing or pairing approvals enabled.
8. Add one scheduled job, like a daily summary, and review its output before letting it send anything externally.