Overview
You already use ChatGPT or Claude through their apps. But what if you had a personal AI assistant that runs on your own infrastructure, uses your own API key, and keeps your conversations private — accessible anywhere through Telegram?
That's exactly what this setup gives you.
What You'll Build
A private Telegram-accessible AI assistant that:
- Responds to your messages using Claude or GPT
- Keeps conversation context across your session
- Stores no data on third-party servers
- Works on any device with Telegram installed
Why Run Your Own?
| Third-party apps | Your own agent | |---|---| | Data used for training | Your API key, your data | | Rate limits on free tier | Scales with your API plan | | Fixed model | Choose any model, switch anytime | | App interface only | Telegram, Discord, or custom |
Setup Steps
1. Create Your Instance
Go to agentclaw.app/signup and create a new agent. This is your personal instance — name it anything.
2. Get a Personal Telegram Bot
Open Telegram and message @BotFather:
/newbot
Follow the prompts to get your bot token.
3. Connect Everything
In your AgentClaw dashboard:
- Integrations → Telegram: paste your bot token
- Integrations → AI Model: add your Anthropic or OpenAI API key
4. Personalize Your Assistant
SYSTEM_PROMPT="You are my personal AI assistant. Be concise, practical, and direct.
I'm a developer who works on [your work context].
Help me with coding, writing, research, and brainstorming."
5. Lock It Down
To keep this private (just for you), configure OpenClaw to only respond to your Telegram user ID. Add this env variable:
ALLOWED_TELEGRAM_IDS=your_telegram_user_id
6. Start Your Instance
Click Start. Your personal AI is live.
Tips
- Model selection: Use Claude Sonnet for the best balance of speed and intelligence
- Context window: OpenClaw maintains conversation history within a session
- Cost: At $0.003 per 1K tokens with Sonnet, heavy personal use is ~$5/month in API costs