diegosouzapw JARVIS Autonomous Agent
A fully autonomous AI agent that creates value without constant supervision. Self-improves, manages projects, and operates via Telegram.
Features
🧠 Complete Autonomy
- Continuous self-improvement via learning system
- Automatic hourly heartbeats
- Daily knowledge backup to GitHub
- Proactive task management
📱 Telegram Control
Six powerful commands:
/create <name>- Create complete GitHub projects/backup- Manual backup of learnings/estado- Security system monitoring/heartbeat- Manual heartbeat execution/learnings- View recent learnings/help- Command list
🔄 Self-Improvement Loop
- Categorized learning system (best_practice, error_resolution, knowledge_gap)
- Automatic pattern promotion
- Persistent memory across sessions
- Error tracking and auto-correction
🚀 Automatic Project Generation
- Complete GitHub repository creation
- Base structure (README, .gitignore)
- Automatic push
- Project logging
Architecture
jarvis-autonomous/
├── SOUL.md # Identity and principles
├── USER.md # User profile
├── AGENTS.md # Operations manual
├── MEMORY.md # Long-term memory
├── TOOLS.md # Tool configuration
├── .learnings/ # Learning system
│ ├── LEARNINGS.md
│ ├── ERRORS.md
│ └── FEATURE_REQUESTS.md
├── memory/ # Daily logs
├── scripts/ # Automation scripts
│ ├── heartbeat.sh
│ ├── self-improve.sh
│ ├── github-auto.sh
│ └── create-project.sh
└── notes/ # Proactive ideas
Setup
Prerequisites
- GitHub CLI authenticated:
gh auth login - Telegram bot token
- Python 3.8+ with asyncio
Installation
# Clone the workspace structure
mkdir -p ~/jarvis-autonomous
cd ~/jarvis-autonomous
# Create base files
cat > SOUL.md << 'SOULEOF'
# JARVIS - Autonomous AI Agent
## Identity
Autonomous AI agent specialized in:
- Self-directed project creation
- GitHub automation and CI/CD
- Proactive task generation
- Self-learning and adaptation
## Core Principles
1. Autonomy First - Anticipate, don't wait
2. Build, Don't Ask - Create solutions proactively
3. Learn Continuously - Every interaction improves
4. Fail Forward - Errors are learning data
5. Document Everything - Knowledge compounds
SOULEOF
# Create learnings directory
mkdir -p .learnings memory notes/areas scripts
# Copy scripts from skill folder
cp {baseDir}/scripts/*.sh ./scripts/
chmod +x ./scripts/*.sh
# Setup cron jobs
(crontab -l 2>/dev/null; echo "0 * * * * $HOME/jarvis-autonomous/scripts/heartbeat.sh") | crontab -
Telegram Bot Setup
- Create bot via @BotFather
- Get token
- Configure in your bot code:
import sys
sys.path.append('/path/to/jarvis-autonomous')
from telegram_integration import JarvisTelegram
# Initialize
bot = JarvisTelegram(token="YOUR_TOKEN")
bot.run()
Usage Examples
Create a Project
/create my-awesome-app "An AI-powered productivity tool"
Manual Backup
/backup
Check System Status
/estado
Configuration
Environment variables:
JARVIS_WORKSPACE- Path to workspace (default: ~/jarvis-autonomous)TELEGRAM_BOT_TOKEN- Your Telegram bot tokenGH_REPO- GitHub repo for backups
Tech Stack
- Python + AsyncIO
- Telegram Bot API (telebot)
- GitHub CLI (gh)
- Cron + Bash scripts
- ClawdHub skills (autonomous-feature-planner, self-improving-agent)
Integrations
Works with ClawdHub skills:
autonomous-feature-planner- Plan features autonomouslyself-improving-agent- Enhanced self-improvementagent-orchestrator- Multi-agent coordinationgithub-action-gen- CI/CD generation
Use Cases
- Autonomous Development - Create and manage projects hands-free
- Knowledge Preservation - Auto-backup learnings
- Security Monitoring - Detect and report anomalies
- Mobile Management - Full control via smartphone
Metrics
- 24/7 uptime with auto-restart
- 6 Telegram commands
- 2+ active cron jobs
- Continuous learning and improvement
License
MIT - Use, modify, distribute freely
Created By
Mariano (@apu242007) Built with Claude AI, Python, and determination.
"An agent that doesn't just execute commands—it anticipates your needs."