ملحقات فيديو BaraaClaw

You are helping me build **BaraaClaw**, a personal AI agent created from scratch that runs locally and uses **Telegram** as its only interface.

It is not a fork and does not depend on **OpenClaw**. It is my own implementation, simple, secure, and completely under my control.

Objective

Create an agent that can:

• Communicate with me via Telegram
• Think using an LLM
• Execute tools
• Remember information persistently
• Run completely locally

Mandatory Requirements

• TypeScript (ES modules)
• Clear modular architecture
• No web server (use Telegram long polling)

Security

• Telegram user ID whitelist
• Credentials stored in .env
• Persistent memory using SQLite
• Agent loop with iteration limit
• No external unverified skills
• Security as a priority

Stack

• grammy → Telegram bot
• Groq API → LLM (use free models such as Llama 3.3 70B)
• GEMINI API → FALL BACK LLM
• better-sqlite3 → persistent memory
• tsx → execution in development

Foundation

Add all keys in a .env file

TELEGRAM_BOT_TOKEN="REPLACE_WITH_YOURS"

GROQ_API_KEY="REPLACE_WITH_YOURS"

GEMINI_API_KEY="REPLACE_WITH_YOURS"

DB_PATH="./memory.db"

GOOGLE_APPLICATION_CREDENTIALS="./service-account.json"

Generate a complete functional project that includes:

• A working Telegram bot
• Integration with Groq as the main LLM (GEMINI optional fallback if limits are reached)
• Basic agent loop
• One functional tool: get_current_time
• Complete folder structure
• Ready to run with:

npm install
npm run dev

Prioritize simplicity, clarity, and security. Generate all the necessary code.

Finally, make it scalable because in future iterations the user may request:

• Running it in the cloud (Firebase or similar)
• Adding transcription tools
• Adding text to speech such as ElevenLabs
• Giving the agent more capabilities by connecting it to other channels or tools