OpenClaw Full Tutorial for Beginners: How to Setup Your First AI Agent (ClawdBot)

OpenClaw (aka ClawdBot) gives you a personal AI agent you control. This guide walks through a safe, practical setup from scratch: choosing the right device, installing OpenClaw, connecting Telegram, adding an API provider, hardening security, enabling web search, and getting useful automation running in the first few days without wasting tokens.

Why run OpenClaw on a dedicated device

Don’t install OpenClaw on your main machine where you keep passwords, documents and logged-in sessions. If the agent gets compromised, it can access files and credentials on that device.

Use a dedicated device such as:

• an old laptop

• a Mac Mini

• a Raspberry Pi (great for lightweight always-on agents)

• a VPS you control (avoid third-party prebuilt images)

One-click deployment images are easy but often lock you into someone else’s configuration. Start from a blank slate so you can change settings and security later.

Install OpenClaw (MVP approach)

Use the provided install command on your chosen device’s terminal. That command typically:

1. checks prerequisites and installs missing packages

2. downloads OpenClaw

3. finalises installation and launches onboarding

My recommendation: install a minimal viable configuration first. Spend a few hours or a day using the agent, learn what you want it to do, then wipe and reconfigure deliberately with tighter settings. No setup is perfectly secure—focus on deliberate limits about who can talk to the agent, where it can act, and what it can touch.

Choose an API provider: Open Router

For flexibility, use Open Router as your gateway for model access. One key gets you access to many providers (OpenAI, Google, Anthropic) and open source models. Open-source model costs can be very low—good for daily-driving tasks—while premium models are available when needed.

Typical steps:

• create an Open Router account

• add a small credit (eg £5–£10) to test

• create an API key and paste it into the OpenClaw onboarding prompt

Telegram bot setup

Telegram is an excellent chat channel for your agent because it’s mobile and always available. To create a bot:

1. open Telegram and message BotFather

2. send the command /newbot

3. give a descriptive bot name and a username ending with bot (use underscores, not dashes)

4. copy the bot token and paste it into the OpenClaw onboarding when prompted

For DM access policies, prefer an allow list. Add only the usernames or user IDs of people who should talk to the agent. To find your Telegram user ID, open your profile and copy the digits from the profile URL.

The first conversation and MD files

When you start a new agent, it will run an onboarding conversation that asks who it is, who you are, and what your goals are. Spend 5–10 minutes on that initial exchange. Keep it simple: let the agent bootstrap its personality and basic identity.

OpenClaw stores important context in markdown (.md) files that the agent can read and update. Key files include:

• bootstrap.md — prompts used on first run

• agent.md — rules and instructions for the agent

• identity.md — agent identity and character

• user.md — information about you (name, timezone, preferences, projects)

Over time you will create domain files like research.md or travel.md. The agent can edit these files when asked, which makes scaling into multiple agents and projects much easier.

Two quick security prompts that double your security

Right after onboarding, run two prompts that massively improve safety. Ask the agent to audit your deployment and to redact sensitive tokens from logs. The agent knows OpenClaw internals and can run health checks and implement fixes.

Example prompt: security audit

Can you run a full security audit of this OpenClaw instance and list any issues or recommended fixes? If you find problems, tell me exactly what you want to change and the commands you will run.

Example prompt: redact sensitive data

Please enable log redaction for sensitive fields. Scan recent conversation metadata for any API keys, user IDs, or tokens and redact them from logs. Report what you found and then apply the redaction rules.

Read the agent’s recommendations. If you agree, reply with a simple confirmation and the agent can apply many fixes automatically. Running those two checks gives you a much safer instance in minutes.

Enable web search: Brave or Perplexity

Give your agent access to web search so it can fetch current information. Brave Search and Perplexity are good options. Brave offers a free tier with a healthy request allowance and easy API key creation.

1. create a Brave account and add an API key

2. paste the key into OpenClaw when asked

3. enable web fetch so the agent can call the API

After setup, test with a simple query and confirm the agent reports success. You can also route Perplexity via Open Router if you prefer.

Set a small, useful automation first

Start with one small automation and iterate. For example: a daily AI news briefing sent at 8am. Run the automation for a few days, then refine scope and add items like weather or YouTube ideas once the agent nails the original task.

Example scheduling prompt

Every day at 08:00 my time, send me a short AI news briefing to my Telegram with 3 headlines, 2 sentence summaries, and one link for deeper reading. Confirm scheduling and show the next run time.

Use MD files to capture ideas and scale slowly

Avoid the temptation to spin up many agents immediately. You’ll waste tokens and time if you don’t know what you need. Instead:

• record voice notes or write quick prompts describing goals, projects and open tasks

• ask the agent to create markdown files—example: research.md, travel.md

• let the agent populate tasks into those files over a few days

• once files are tidy, ask the agent to spawn specialised agents based on those MD files

This approach builds a small set of strong context files that let new agents start with clarity and saves tokens.

Quick setup checklist

• Use a dedicated device or VPS you control (avoid prebuilt one-click images).

• Install OpenClaw and pick manual onboarding with minimal options first.

• Use Open Router for model access and paste the API key during onboarding.

• Create a Telegram bot with BotFather and paste the token into OpenClaw.

• Set an allow list for DM access so only authorised users can talk to the agent.

• Run a security audit prompt and a redaction prompt immediately.

• Enable web search (Brave or Perplexity) and test queries.

• Start one small automation (eg daily AI news) and iterate for a few days.

• Use MD files to capture projects, then spin up additional agents from those files.

Final tips and next steps

Move slowly during the first few days. Learn the agent’s strengths and limits, then be deliberate about which APIs, permissions and devices the agent can access. Use markdown files to centralise memory and tasks before expanding into multiple agents.

Once you have a stable, secure base, you can build agent teams, a command centre for multi-agent workflows, and richer automations. Start small, secure early, and iterate.