Autonomous self-healing agent for decentralized cloud deployments on Akash Network.

AkashGuard monitors services deployed on Akash Network, diagnoses failures using AI, and autonomously recovers them — zero human intervention required. When a provider goes down, AkashGuard detects the failure, reasons about the best recovery action via LLM, and redeploys to a new provider automatically.

Built for the Open Agents Hackathon (Feb 25, 2026 — Yes SF, San Francisco).

Monitor Loop (every 30s)
  ├─ Health check all registered services
  ├─ Record metrics (status code, response time, errors)
  ├─ If service unhealthy (3+ consecutive failures):
  │   ├─ LLM Diagnosis (Llama 3.3 70B via AkashML)
  │   │   → Analyzes health history, recommends action
  │   ├─ Decision: redeploy / wait / scale
  │   ├─ If redeploy (confidence > 70%):
  │   │   ├─ Close current lease on Akash
  │   │   ├─ Redeploy via Akash Console API
  │   │   ├─ Wait for new provider + lease acceptance
  │   │   └─ Verify recovery with health check
  │   └─ Send Telegram alert
  └─ Stream events to dashboard via SSE

• Autonomous Recovery — Detects failures and redeploys to new providers without human intervention
• AI-Powered Diagnosis — Uses Llama 3.3 70B to analyze health data and make intelligent recovery decisions
• Real-Time Dashboard — Live monitoring via Server-Sent Events with service status, decision history, and provider tracking
• Telegram Alerts — Instant notifications on service down/recovery events
• Post-Recovery Cooldown — 120s stabilization period prevents recovery thrashing
• Provider Scoring — Tracks provider reliability over time
• Graceful Degradation — Never crashes on partial failures; degrades intelligently
• Demo Mode — Simulated failure injection for live demonstrations

akashguard/
├── agent/                      # AkashGuard autonomous agent
│   ├── main.py                 # Agent loop orchestrator
│   ├── api.py                  # FastAPI server + SSE dashboard
│   ├── health_checker.py       # HTTP health monitoring
│   ├── llm_engine.py           # LLM-powered diagnosis engine
│   ├── recovery_engine.py      # Akash Console API recovery
│   ├── database.py             # SQLite persistence layer
│   ├── event_bus.py            # Asyncio event queue (SSE)
│   ├── notifier.py             # Telegram alert integration
│   ├── config.py               # Pydantic settings
│   └── static/
│       └── dashboard.html      # Real-time monitoring UI
│
├── chatbot/                    # Monitored service (Flask chatbot)
│   ├── app.py                  # Flask + AkashML Llama chat
│   ├── Dockerfile
│   └── requirements.txt
│
├── deploy/                     # Akash deployment configs
│   ├── agent-sdl.yaml          # AkashGuard SDL
│   └── chatbot-sdl.yaml        # Chatbot SDL
│
├── misc/                       # Testing & debugging scripts
├── data/                       # SQLite database (gitignored)
├── .env.example                # Environment variable template
└── README.md

• Python 3.11+
• API keys: AkashML, Akash Console, Telegram (optional)

# Clone
git clone https://github.com/xploit007/akashGuard.git
cd akashGuard

# Install dependencies
pip install -r requirements.txt

# Configure environment
cp .env.example .env
# Edit .env with your API keys

# Agent + API (monitoring + dashboard)
$env:AGENT_AUTO_MONITOR = "true"
uvicorn agent.api:app --host 0.0.0.0 --port 8000

# API only (no active monitoring)
$env:AGENT_AUTO_MONITOR = "true"
uvicorn agent.api:app --host 0.0.0.0 --port 8000

# Build and push Docker images
docker build -t DOCKER_HUB_USERNAME/akashguard-agent:latest agent/
docker push DOCKER_HUB_USERNAME/akashguard-agent:latest

docker build -t DOCKER_HUB_USERNAME/akashguard-chatbot:latest chatbot/
docker push DOCKER_HUB_USERNAME/akashguard-chatbot:latest

# Deploy via Akash Console or CLI using SDL files in deploy/

The API starts with no services registered. To have the agent monitor your chatbot (or another service), register it once:

1. Check services (optional — will show 0 at first):

   Invoke-RestMethod http://localhost:8000/api/services

2. Register the chatbot (initializes DB and adds the service with its SDL):

   python -c "from pathlib import Path; from agent.database import init_db, add_service; init_db(); sdl=Path('deploy/chatbot-sdl.yaml').read_text(); print(add_service('chatbot','http://localhost:5000/health',sdl))"

3. Verify — list services again; you should see the registered service:

   Invoke-RestMethod http://localhost:8000/api/services

After this, the agent will monitor the registered service and the project is ready to use.

See .env.example for the full list. Key variables:

1. Show dashboard — Services green, agent monitoring
2. Kill a deployment — Simulate provider failure
3. Watch AkashGuard — Agent detects failure, LLM diagnoses, auto-redeploys
4. Service returns — New provider, new lease, service back online
5. Telegram alert — Real-time notification of recovery

MIT