🏠 Graham's AI Orchestrator

The Autonomous Multi-Agent Brain for your Smart Home.

🚀 Alpha Release (v0.10.0): Now seeking testers!

The AI Orchestrator transforms your Home Assistant from a collection of manual toggles and rigid automations into a dynamic, thinking ecosystem. It deploys Autonomous AI Agents that reason about your home's state, understand your intent, and execute actions intelligently using a built-in RAG Knowledge Base and Safe Tool Execution Engine.

(Note: Dashboard images are representative)

---

🌟 Why use this?

🧠 Real Reasoning, Not Scripts

Instead of writing complex YAML for "If motion > 5mins AND sun is down", just tell an agent:

"You are the Security Guard. If the front porch is occupied for more than 5 minutes at night, turn on the floodlight and notify me."

🗣️ Natural Language Chat

A Floating Chat Assistant lives in your dashboard. Talk to your house naturally:

"It's movie night. Get the living room ready."
"Who left the garage door open?"

🔧 No-Code Agent Factory

Don't know how to prompt? The Architect AI will interview you and build the perfect agent for your needs automatically. It even surfaces Smart Suggestions based on your devices!

📚 RAG Knowledge Base

Feed your AI manuals, PDF guides, or specific house rules. The agents read them before acting.

"Don't turn on the dishwasher if the 'Clean' sign is flipped."

🎨 AI Visual Dashboard

Experience high-fidelity, skeuomorphic visualizations generated in real-time. The AI Visual Dashboard allows you to dynamically request and rebuild your entire home interface using natural language.

• No YAML, No Coding: Because the system has deep access to your Home Assistant Entity Registry, it builds the dashboard autonomously. You don't need to manually configure cards or write a single line of YAML.
• Dynamic Design: Just tell the Architect Agent: "Make a dashboard focused on energy usage with a dark oceanic theme" or "Build a 3-column view for my heating and security" and watch it regenerate in seconds.
• Advanced LLM Support: Leverages both local (Ollama) and cloud-based (Google Gemini) advanced LLMs for stunning, context-aware UIs.
• Quick Trigger: Use the FAB quick actions for instant generation or access via the dedicated sidebar button.

---

📦 How to Install

1. Add Repository: Copy this URL:
   https://github.com/ITSpecialist111/HASS-AI-Orchestrator

   Go to Home Assistant > Settings > Add-ons > Add-on Store > Repositories (3 dots top right) and add it.

2. Install "AI Orchestrator": Find it in the list and click Install.

3. Configure:

   • Ollama Models: Ensure you have pulled the required models on your Ollama server:

     ollama pull deepseek-r1:8b
     ollama pull mistral:7b-instruct
     ollama pull nomic-embed-text  # Mandatory for RAG Knowledge Base

   • Tokens: Create a Long-Lived Access Token in your Profile.

4. Start & Explore: Open the Web UI to see your agents thinking in real-time!

---

🔒 Privacy & Safety

• 100% Local: Works with your local Ollama instance. No data leaves your network.
• Safe Execution Engine (v0.9.9):
  • Configurable Safety: All security settings (Allowlists, Blocked Domains, Impact Services, Temp Limits) are now editable via the Add-on Configuration tab in Home Assistant.
  • Domain Allowlist: AI only has access to safe domains (light, switch, climate, etc.) by default.
  • Critical Guard: Dangerous domains (shell_command, hassio, etc.) are explicitly blocked.
  • Approval Queue: High-impact actions (unlocking doors, disarming alarms) require manual approval.
• Dry Run Mode: Test your agents efficiently without them actually touching your devices until you trust them.
• Anonymized Telemetry: We use ChromaDB for local vector storage. By default, it sends basic, anonymized usage statistics (client start, collection creation). No personal house data is ever collected. You can disable this by setting CHROMA_TELEMETRY_EXCEPT_OPT_OUT=True in your environment.

---

⚡ Agent Speed & Real-Time Monitoring

Agents are rate-limited by default to a 120-second decision interval to save CPU and LLM resources.

• Custom Intervals: Set individual intervals in your agents.yaml or directly in the Dashboard.
• Real-Time Monitoring: Want an agent to react instantly? Set its interval to 5-10 seconds. We call this "Real-Time" mode—ideal for lighting or security presence.
• Speed Note: The actual reaction time is limited by your LLM inference speed. A dedicated GPU server is recommended for "Real-Time" behavior.

---

⚠️ Requirements

• Home Assistant OS (or Supervised)
• Ollama Server (Running locally or on another machine in your network)
• Hardware: Raspberry Pi 4 (8GB) or NUC recommended for fluid dashboard performance. External LLM server recommended for AI speed.

---

Read Full Documentation