How To Use ClawdBot: Building Your Own 24/7 AI Assistant

A Practical Guide to Setting Up, Configuring, and Mastering Your Personal AI That Actually Does Things

~64.000 Stars in a couple of days can't be wrong...

Introduction: The AI Assistant We Were Promised

Listen, I get it. We've been sold on the promise of AI assistants for years. Siri, Alexa, Google Assistant... they all have one thing in common: they're pretty useless.

They can't remember what you told them yesterday. They can't read your emails and automatically create tasks. They definitely can't check you in for flights or manage your smart home without an elaborate setup. They're reactive, not proactive. They sit there waiting for commands.

Enter ClawdBot.

ClawdBot isn't ChatGPT. It's not a web interface you check in between tabs. ClawdBot is an actual AI agent that lives on your computer (or a $5 VPS), connects to the apps you already use (Telegram, Discord, WhatsApp, Slack, you name it), remembers your preferences, and, here's the kicker, can actually do things on your behalf. It can read files, run scripts, browse the web, fill out forms, send emails. It's the AI assistant we were supposed to get in 2011, except it's finally here in 2026.

And the best part? It's free, open-source, and runs completely under your control.

In this guide, I'm going to walk you through everything: how to install it locally or on AWS, how to connect it to Telegram (the most popular option), how to actually use it with real prompts and examples, and the kinds of workflows that transform how you work.

Let's go.

1. Understanding ClawdBot: Why This Is Different

Before we get into the installation weeds, let me explain why ClawdBot is fundamentally different from every other AI assistant you've tried.

What Makes It Special?

Most AI assistants are stateless. You send a message, they respond, and that's it. The conversation exists in isolation. They don't know who you are beyond the current chat. They can't remember your preferences, your schedule, your workflows. Every conversation starts from scratch.

ClawdBot changes that. Here's how:

Persistent Memory: ClawdBot remembers you. It learns your context over time. You can tell it your preferences once, and it carries them forward across every future conversation. If you tell it you prefer email in Markdown format, it will format every email in Markdown forever. If you mention you get anxious about 8 AM meetings, it will try to avoid scheduling you for 8 AM.

Full System Access: ClawdBot has read and write access to your files, your system, your scripts. You can tell it "analyze this folder of customer feedback and create a summary" and it will do it. You can say "check if the CI pipeline passed" and it will SSH into your server and check. You can ask it to "generate a weekly report from my database" and it will query the database, process the results, format it nicely, and send it to you via Telegram.

Proactive Capabilities: This is the big one. Most assistants are reactive. ClawdBot can reach out to you first. It can send you a morning briefing with your calendar, your top tasks, and any urgent emails. It can check for specific events and alert you. It can run scheduled tasks. It's like having an assistant who works 24/7 and doesn't need coffee.

Works Everywhere: ClawdBot connects to every messaging app. Telegram, Discord, WhatsApp, Slack, Signal, iMessage, Teams. You have one AI assistant, and you can talk to it from wherever you are. One inbox for all conversations.

Costs Way Less: A basic ClawdBot setup costs around $20/month for Claude Pro (if you use the paid API). Compare that to ChatGPT Plus ($20/month) + assistant services (varies wildly). But here's the thing: if you run ClawdBot on a $5 Hetzner VPS or AWS Free Tier, your compute costs are almost nothing. You're really just paying for the LLM API.

Why Open Source Matters

ClawdBot is MIT licensed and open-source. That means:

Your data stays yours. It's not being collected, analyzed, or used to train other models.
You own the entire system. No vendor lock-in. If the original maintainers stop working on it, the community can fork it.
You can modify it. If you want to add a feature or change how it works, you can.
Security is transparent. Anyone can audit the code. No hidden backdoors.

This is critical if you're going to give an AI assistant access to your files, your email, your calendar. You need to know exactly what it's doing.

2. Installation: Three Paths (Pick Your Own Adventure)

There are three main ways to run ClawdBot. Pick based on your comfort level and your use case.

Path 1: Local Installation (~15 minutes)

Best for: Developers, people comfortable with the terminal, or anyone who wants to try it out before committing to a server.

Requirements:

Node.js 22+ (if you don't have it, install from nodejs.org)
2GB RAM minimum (an old laptop works fine)
Any OS: macOS, Linux, Windows (WSL2 recommended)

Steps:

Open your terminal and run the one-line installer:

curl -fsSL https://clawd.bot/install.sh | bash

That's it. The installer handles everything: downloads ClawdBot, sets up the directory structure, installs dependencies.

Then run:

exec bash

This reloads your shell and adds ClawdBot to your PATH.

Start the onboarding wizard:

clawdbot onboard --install-daemon

The wizard will ask you:

Whether you want local or cloud hosting (pick "local" for now)
Which LLM provider you want (Anthropic/Claude recommended, but OpenAI works too)
Your API key (get one free or paste your existing key)
Which messaging channels you want to enable (we'll set up Telegram)

Once done, start the gateway:

clawdbot gateway --port 18789

Open your browser to http://127.0.0.1:18789 and you'll see your dashboard. ClawdBot is now running locally on your machine.

Pros: Fast, no server costs, full control, good for learning.

Cons: Only works when your computer is on. If you turn off your Mac, Telegram messages won't be answered until you turn it back on.

Path 2: AWS Free Tier (~20 minutes)

Best for: People who want 24/7 uptime but don't want to pay much.

AWS offers a generous free tier: 750 hours/month of t2.micro or t3.micro EC2 instances. That's enough to run ClawdBot continuously for free for the first year.

Steps:

Create an AWS account if you don't have one. (Go to aws.amazon.com, click "Create an AWS Account", follow the prompts.)
Launch an EC2 instance:
- Go to EC2 Dashboard → "Launch Instances"
- Select Ubuntu 24.04 LTS
- Choose instance type: t2.micro (Free Tier eligible)
- Create a new key pair (download the .pem file and keep it safe)
- In security group, allow SSH (port 22) and HTTP (port 80) from your IP
- Launch
Connect to your instance:
- Go back to EC2 Instances, find your instance, click "Connect"
- Use the browser-based terminal (easiest option)
Run the installer:

curl -fsSL https://clawd.bot/install.sh | bash
exec bash
clawdbot onboard --install-daemon

Configure and start:

clawdbot gateway --bind 0.0.0.0 --port 80

Your ClawdBot is now running on AWS and accessible from anywhere.

Important security note: The setup wizard will ask about security. For production, enable authentication. Don't expose your dashboard to the internet without protection.

Pros: Always on, cheap/free first year, scales easily if you need more power later.

Cons: Slightly more complex setup, need to manage AWS console.

Path 3: Hetzner VPS (~5/month, 15 minutes)

Best for: People who want a simple, cheap, always-on setup with good uptime.

Hetzner is a datacenter company with transparent pricing. A basic VPS is $5.41/month. That's about the cost of a coffee.

Steps:

Go to hetzner.com/cloud and sign up.
Create a new project.
Create a new server:
- OS: Ubuntu 24.04 LTS
- Server type: CX12 (2 GB RAM, good enough)
- Location: Pick the one closest to you
- Create
You'll get a password. SSH into the server:

ssh root@YOUR_SERVER_IP

Create a new user (for security, don't run ClawdBot as root):

adduser clawd
usermod -aG sudo clawd
su - clawd

Install ClawdBot:

curl -fsSL https://clawd.bot/install.sh | bash
exec bash
clawdbot onboard --install-daemon

Start the gateway:

clawdbot gateway --bind 0.0.0.0 --port 80

Done. Your ClawdBot is now running 24/7 on a $5/month VPS.

To access the dashboard from your laptop, use SSH tunneling:

ssh -L 18789:127.0.0.1:18789 clawd@YOUR_SERVER_IP

Then open http://127.0.0.1:18789 locally.

Pros: Cheap, simple, transparent pricing, excellent support.

Cons: One more service to manage, slightly less integrated with your system than local.

My Recommendation: Start with local (Path 1) to learn. Once you're comfortable and want 24/7 uptime, migrate to Hetzner (Path 3). It's the best bang for your buck, and the migration takes about 15 minutes.

3. Connecting Telegram: Your Interface to Everything

Once ClawdBot is installed, you need an interface. Telegram is the most popular choice (and my recommendation), so let's set that up.

Why Telegram?

Private and secure: End-to-end encryption available.
Bot API is solid: Reliable, well-documented, no weird rate limits.
Works everywhere: iOS, Android, web, desktop.
You probably already use it: If not, install it. It takes 2 minutes.

Setting Up Telegram Bot (5 minutes)

Create a bot token:
- Open Telegram and search for @BotFather
- Send the command /newbot
- BotFather will ask for a name (e.g., "MyClawdBot") and a username (must end in "bot", e.g., "my_clawd_bot")
- BotFather gives you a token. Copy it. Keep it secret. (If you leak it, anyone can control your bot. Go back to @BotFather and regenerate.)

Configure ClawdBot:

Option A (easiest): Environment variable

export TELEGRAM_BOT_TOKEN=123456:ABC-DEF1234ghIkl-zyx57W2v1u123ew11
clawdbot gateway

Option B (persistent): Add to config

Create ~/.clawdbot/config.json and add:

{
  "channels": {
    "telegram": {
      "enabled": true,
      "botToken": "YOUR_TOKEN_HERE"
    }
  }
}

Test it:
- Open Telegram, search for your bot username, click "Start"
- Type a message: "Hello"
- ClawdBot should respond

If it doesn't respond, check logs:

clawdbot logs --follow

Advanced Telegram Configuration

ClawdBot has a ton of Telegram options. Here are the most useful:

DM Policy (who can message you)

"dmPolicy": "pairing"

Options:

"pairing": Unknown users get a pairing code. You approve them first. (Recommended for security)
"allowlist": Only people you explicitly allow
"open": Anyone can message
"disabled": No DMs

Group Management

By default, ClawdBot works in DMs but not in groups. To enable groups:

"channels": {
  "telegram": {
    "groups": {
      "*": { "requireMention": true }
    }
  }
}

This allows ClawdBot to respond in any group, but only if mentioned with @your_bot_name.

If you want it to respond in specific groups without mention:

"groups": {
  "-1001234567890": { "requireMention": false }
}

(To find your group's chat ID: send any message in the group, then run clawdbot logs --filter "chat" and look for the number.)

Voice Support

ClawdBot can send voice messages (with ElevenLabs TTS) and will respond to voice messages:

"channels": {
  "telegram": {
    "voiceEnabled": true,
    "ttsProvider": "elevenlabs"
  }
}

(You'll need an ElevenLabs API key, which is free to start with.)

4. Real Examples: Prompts and Actions That Actually Work

Okay, now the fun part. Let's talk about what you actually do with ClawdBot once it's running.

Example 1: Morning Briefing (The Proactive Use Case)

The Setup: You want ClawdBot to send you a message every morning at 8 AM with:

Your calendar for today
Your top 3 tasks from Notion
Any urgent emails
Weather for your location

The Prompt:

You'd set up a "heartbeat" (scheduled task) in ClawdBot's config:

"agents": {
  "main": {
    "heartbeats": [
      {
        "name": "morning-briefing",
        "schedule": "0 8 * * *",
        "prompt": "Send me a Telegram message with: 1) Today's calendar events, 2) My top 3 tasks from Notion (check the Notion database), 3) Any emails from VIPs (marked as starred), 4) Weather for Bucharest. Format it nicely with emojis. Morning briefing should take 2-3 lines max."
      }
    ]
  }
}

What Happens:

At 8 AM every day, ClawdBot wakes up
It reads your calendar API (you need to configure this once)
It queries your Notion database
It checks your email inbox for starred messages
It fetches the weather
It compiles everything into a nice summary
It sends it to your Telegram DM

You get something like:

🌅 Morning Briefing - Jan 26

📅 Calendar:
• 9:30 AM - Team Standup (Zoom)
• 2:00 PM - Client Call

✅ Top 3 Tasks:
• Finish ClawdBot article (HIGH)
• Review PR for payment feature
• Fix bug in login flow

📧 Important Emails:
• From: CEO - "Q1 Planning" (starred)

🌤 Weather: 12°C, Partly Cloudy. Bring a jacket!

Every morning. No manual work.

Example 2: Inbox Zero Automation (The Background Task)

The Use Case: Your inbox is a disaster. 200+ unread emails. You want ClawdBot to:

Read all emails from the past 24 hours
Categorize them (work, personal, spam, newsletters)
Create tasks in Notion for anything actionable
Summarize newsletters and file them
Send you a summary

The Prompt:

"I'm drowning in emails. Read all my unread emails from the past 24 hours. 
For each one:
1) If it's actionable (needs a response or action), create a task in my Notion database called 'Email Follow-ups' with the sender, subject, and what I need to do.
2) If it's a newsletter, summarize it in 1-2 sentences and move it to a folder called 'Newsletters'.
3) If it's spam/promotional, trash it.
4) If it's important (from my boss, clients, or family), star it and send me a Telegram notification.

Then send me a summary: how many emails processed, how many actions created, any VIPs I need to reply to."

What Happens: ClawdBot:

Connects to your email (Gmail, Outlook, whatever)
Reads 24 hours of emails
Classifies each one
Creates tasks in Notion for actionable items
Sends a summary to Telegram

Result: Your inbox is organized, and you have a task list of what actually needs your attention.

Example 3: Form Filling and Data Extraction (The Browser Automation)

The Use Case: Every week, you need to fill out a timesheet on your company's website. It's tedious. You also need to extract last week's data from an analytics dashboard and put it in a spreadsheet.

The Prompt:

"I need you to:
1) Go to timesheet.ourcompany.com and log in with my credentials (I'll store them in your config).
2) Fill in this week's hours: Monday-Thursday 8h, Friday 4h.
3) Set project codes: 'PROJECT-123' for first 3 days, 'PROJECT-456' for Thursday, 'ADMIN' for Friday.
4) Submit the timesheet.
5) Then go to analytics.ourcompany.com and extract the 'Revenue' and 'Users' metrics from last week.
6) Send me a Telegram message with the numbers and a simple chart."

What ClawdBot Does:

Opens browser (Playwright headless)
Logs in
Fills form fields
Submits
Navigates to analytics
Extracts data
Sends summary to Telegram

You spend 0 minutes on a task that would normally take 15 minutes. Weekly.

Example 4: Code Review Automation (The Developer Workflow)

The Use Case: Your GitHub PR notifications are out of control. You want ClawdBot to:

Check for PRs assigned to you or requesting your review
Summarize each PR (files changed, lines added/removed, description)
Flag potential issues (large diffs, unusual patterns)
Run your linter/test suite on the code
Send you a digest

The Prompt:

"Check GitHub for PRs assigned to me or requesting my review.
For each PR:
1) Clone the repo and check out the branch
2) Run 'npm run lint' and 'npm run test'
3) If tests fail, tell me which ones and why
4) Count the files changed and lines added/removed
5) Look for security issues or obvious bugs in the diff
6) Create a summary with: title, author, # of files, # of lines, test status, any issues

Send me the digest sorted by priority (failing tests first, then large diffs, then small fixes)."

What Happens: You get a digest like:

🔍 PR Digest - 3 PRs waiting

❌ HIGH PRIORITY:
PR #445: "Add payment feature" by @alex
- 12 files changed, 450 lines added
- TESTS FAILING: payment_test.js (3 failures)
- Issue: Missing error handling in crypto validation

⚠️ MEDIUM PRIORITY:
PR #442: "Refactor dashboard" by @sarah
- 8 files changed, 200 lines modified
- Tests passing ✅
- Note: Large diff, may want to review in chunks

✅ LOW PRIORITY:
PR #440: "Fix typo in README" by @mike
- 1 file changed, 1 line
- Tests passing ✅

5. The Real-World Impact: What Changes When You Do This

I want to be honest about something. Most AI tools overpromise and underdeliver. They're nice to have, but they don't fundamentally change how you work.

ClawdBot is different because it actually does things. Let me give you real scenarios from people using it:

Scenario 1: The Developer (30 minutes saved per day)

Before ClawdBot:

8:00 AM: Check email manually
8:15 AM: Triage PRs on GitHub
8:30 AM: Check CI/CD pipeline status
9:00 AM: Finally start actual work

After ClawdBot:

8:00 AM: Check Telegram. ClawdBot sent you a briefing with PRs needing review, failing tests, and this morning's meetings. You scan it while drinking coffee.
8:05 AM: Start actual work (the important stuff)

That's 25-30 minutes a day. 2+ hours a week. Over 100 hours a year.

Scenario 2: The Manager (5+ hours saved per week)

Before ClawdBot:

Monday morning: Manually check everyone's status
Tuesday: Compile weekly report from various sources
Wednesday: Update project tracking sheet
Thursday: Prepare metrics for leadership review
Friday: Create summary of what shipped

After ClawdBot:

Set up ClawdBot once to pull data from Jira, GitHub, Slack
Every Friday at 4 PM, it sends you a complete report: PRs merged, bugs fixed, velocity metrics, blockers, team sentiment
You spend 10 minutes reviewing instead of 5 hours compiling

Scenario 3: The Designer/Content Creator (automation of the "boring bits")

Before ClawdBot:

Design something
Export assets manually
Resize for different platforms (Instagram, Twitter, LinkedIn)
Upload manually
Write different copy for each platform
Schedule posts

After ClawdBot:

Design once
Tell ClawdBot: "I've designed a post. Resize it for Instagram (1080×1350), Twitter (1024×512), and LinkedIn (1200×628). Write copy for each platform (Instagram: casual, Twitter: witty, LinkedIn: professional). Schedule for tomorrow at 9 AM on each platform."
ClawdBot does it in 2 minutes

You've saved an hour on something that didn't need your intelligence. That hour can go to actual creative work.

6. Security & Privacy: The Important Stuff

Before you give ClawdBot access to your entire digital life, let's talk about security.

What ClawdBot Can't Do (By Design)

It doesn't send your data to the cloud (unless you choose to). Everything runs locally or on your server.
It doesn't train on your data. Your files, emails, conversations, they're not used to improve Anthropic's models.
It can't phone home. ClawdBot is sandboxed. It can't exfiltrate data unless you explicitly give it a tool to do so.
It's not monitoring you. It doesn't send screenshots, key logs, or any creepy telemetry.

What You Should Do

Use strong authentication. If ClawdBot is accessible from the internet, put it behind a password.

"gateway": {
  "auth": {
    "enabled": true,
    "password": "some_strong_password_here"
  }
}

Limit permissions. ClawdBot doesn't need access to your entire system. Create a limited user account for it.

useradd --shell /bin/false --home-dir /opt/clawd clawd

Store credentials safely. If you're giving ClawdBot your email password or API keys, store them in environment variables or a secure config file, never in plain text.
Audit logs regularly. Check what ClawdBot is actually doing.

clawdbot logs --filter "action" --since "7 days ago"

Sandbox file access. If you don't need ClawdBot accessing your entire home directory, limit it to a specific folder.

"agents": {
  "main": {
    "workspace": "/opt/clawd/workspace"
  }
}

The Bottom Line

ClawdBot is open-source. You can read the code. You can audit it. You control it completely. That's the opposite of closed-source AI services that collect data and train models on your inputs.

Is there a risk? Sure. Any system with access to your files/email has risk. But you control the risk. You decide what gets access to what.

7. Advanced: Skills, Custom Tools, and Self-Improving Agents

Once you're comfortable with the basics, you can do some genuinely wild stuff with ClawdBot.

Skills (Extending Capabilities)

ClawdBot has a "skills" system. You can create a skill, basically, a set of instructions or tools that the agent can load on demand.

Example skill file (~/.clawd/skills/my_research_skill.md):

# Research Skill

When I ask you to "research X", you should:
1. Use the Brave Search API to find 5 recent articles
2. For each article, extract: title, source, date, key points (3 bullets)
3. Summarize common themes across all 5 articles
4. Rate confidence (low/medium/high) based on consistency
5. Send me the summary via Telegram

**Tool**: Brave Search API (configured in config.json)
**Cost**: 1 search credit per research (~$0.001)

Then you can just tell ClawdBot: "Research the latest developments in AI agents" and it will load the skill, execute it, and send you a summary.

Custom Tools

You can give ClawdBot access to custom tools. For example, if you have an internal service or API that ClawdBot doesn't know about, you can define it:

"tools": {
  "custom_api": {
    "name": "Our Internal Analytics API",
    "description": "Fetches revenue, user, and retention metrics",
    "url": "https://api.our-company.com",
    "auth": "Bearer YOUR_API_KEY",
    "methods": [
      {
        "name": "get_metrics",
        "endpoint": "/metrics",
        "params": ["date_from", "date_to", "metric_type"]
      }
    ]
  }
}

Now ClawdBot can query your internal API directly.

The Self-Improving Loop

Here's where it gets really interesting. You can set ClawdBot to improve itself through Discord or Telegram feedback.

Set up a feedback loop:

"feedback": {
  "enabled": true,
  "channel": "telegram",
  "selfImprove": true
}

Now when you give feedback ("That was wrong, try this instead"), ClawdBot can:

Store your feedback
Analyze what went wrong
Create a new skill or modify an existing one
Test it on the feedback scenario
Deploy the improvement

You're literally teaching your AI assistant how to be better. Over time, it gets better at understanding your specific needs.

8. Common Issues & Troubleshooting

ClawdBot isn't responding to Telegram messages

Check:

clawdbot logs --follow

Look for errors. Common issues:

Bot token wrong → regenerate via @BotFather
Firewall blocking Telegram API → unlikely but check your network
ClawdBot service not running → run clawdbot gateway again

Telegram messages are slow to respond

Check if your LLM provider (Anthropic, OpenAI) is having issues
Check your internet connection
If running locally, maybe your computer went to sleep → keep it awake or move to server

"Permission denied" errors

You're probably running ClawdBot as the wrong user. Run:

clawdbot status

Check what user it's running as. If it needs access to your home directory, make sure that user has permissions.

My email/API credentials aren't working

Make sure they're stored correctly:

clawdbot config --list

Check for typos. Also check if your email provider requires an app-specific password (Gmail, Microsoft) rather than your main password.

I accidentally exposed my bot token on GitHub

No problem, it happens. Go to @BotFather and run /getmycommands, find your bot, regenerate the token. Update your config. Done.

9. The Path Forward: What's Possible

We've covered the basics, but honestly, we're still in the early innings of what ClawdBot can do.

Right now, people are using it for:

Inbox management: Zero inbox achieved
Calendar optimization: Automatic meeting scheduling and prep
Code review: Automated PR analysis and testing
Content creation: Bulk image resizing, copy writing, scheduling
Data analysis: Extracting insights from databases and APIs
Personal research: Daily briefings on topics that matter to you
Smart home: One interface for all your devices
Flight check-ins: Automated check-in reminders and execution

The pattern is: anything repetitive, anything that doesn't require creative judgment, anything that wastes your time; ClawdBot can probably do it.

What's Coming

The creator, Peter Steinberger, is actively developing ClawdBot. On the roadmap:

Better vision capabilities (understanding screenshots and images more deeply)
Voice wake-word support (say "Hey Clawd" and it listens)
Deeper integrations with more platforms
Better model reasoning (for more complex tasks)
Community-driven skills marketplace

10. Getting Started: Your Next Steps

Okay, I've thrown a lot at you. Here's what to actually do:

Install locally (takes 15 min). Use Path 1 above. Try it out.
Connect Telegram (takes 5 min). Get a bot token, set it up.
Send a test message. Something like: "What's 2+2? Also, what's today's date?"
Try a simple task. Something like: "Write me a poem about coffee" or "What's the weather like in [your city]?"
Then try something useful. Schedule a morning briefing. Set up inbox automation. Create a research task.
Once you're comfortable, migrate to a server if you want 24/7 uptime. Hetzner is $5/month and takes 15 minutes.
Join the community. ClawdBot has a Discord (discord.gg/clawd). People are sharing workflows, skills, and ideas.

Resources

Official docs: https://docs.clawd.bot
GitHub: https://github.com/clawdbot/clawdbot
Skills marketplace: https://clawdhub.com/skills
Creator: Peter Steinberger (@steipete on Twitter)
Discord community: discord.gg/clawd

Final Thoughts

We've been waiting for an AI assistant that actually does things. Not one that chats with you. Not one that gives you cute poems. One that actually understands your life, remembers your preferences, and handles your repetitive work.

ClawdBot is that assistant.

The best part? It's open-source, it's cheap (or free for the first year on AWS), and it's completely under your control. No vendor lock-in. No data collection. No surprises.

Start small. Try it locally. See what it can do for you. I bet you'll be surprised.

Questions? Want to share your ClawdBot setup?

Find me on x.com/axelut. I'd love to hear what you're automating.

P.S. If you create something cool with ClawdBot, share it. The community is small right now, but it's growing fast. And honestly, the most interesting applications we haven't even thought of yet will come from people like you experimenting and sharing what works.

The future of AI assistants isn't in Silicon Valley building closed-source systems. It's in your terminal, doing your work, remembering who you are, and actually getting things done.