docs

Documentation

Thane is an autonomous AI agent that connects language models to Home Assistant, giving AI grounded awareness of physical reality. These docs cover what Thane is, how it thinks, and how to run it.

New here? Start with Your First Thane — a complete onboarding guide for Home Assistant users.

---

Understanding Thane

What is this and how does it think?

• Philosophy — Why Thane exists, why Home Assistant is foundational, and why privacy is structural
• Architecture — System design, component map, technology choices
• The Agent Loop — The core reasoning cycle: context assembly, tag activation, planning, delegation, response
• Delegation — How the orchestrator model plans and local models execute at zero cost
• Memory — Semantic facts, conversation history, session archives, and how the agent learns
• Context Layers — How the system prompt is assembled from persona, talents, knowledge, and session state
• Document Roots — How paths: becomes named local document collections that Thane can keep track of
• Trust Architecture — Safety through structural enforcement, not prompt compliance
• Glossary — Canonical definitions for Thane-specific terminology

Operating Thane

How do I run this?

• Getting Started — Build, configure, and run (quick reference)
• Your First Thane — Complete guide: hardware, models, deployment, and building a relationship with your agent
• Home Assistant — Connect Thane as your HA conversation agent
• MQTT — Broker setup, telemetry entities, and wake subscriptions
• Configuration — Config guide organized by concern
• Routing Profiles — Model selection presets (thane:latest, thane:premium, thane:ops, etc.)
• Hardware Requirements — Platform support, production observations, minimum specs
• Deployment — Service installation for macOS and Linux
• Release Engineering — Preferred macOS Tahoe workflows for GitHub releases and pkg-based live-host deploys

Reference

What exactly does X do?

• Tools — All 80+ native tools by category
• CLI — Commands, flags, and config discovery
• API & Endpoints — Ports, protocols, web dashboard, CardDAV
• Event Sources — Everything that can wake the agent loop

Extending Thane

• OpenClaw Compatibility — Run an OpenClaw-style workspace agent through Thane's thane:openclaw profile

Developer Internals

• Prompt & Tool Pruning Strategy — Tracking doc for shrinking the always-on prompt and moving domain doctrine behind tagged context
• Model-Facing Context — Philosophy and conventions for writing data consumed by model loops
• Model-Facing Tools — Guidance for naming, shaping, and erroring internal tools
• Dashboard Graph Visual Grammar — Visual language for the dashboard's force-directed graph

Project

• History — Timeline from first commit to v0.9.0
• Contributing — Development workflow and guidelines
• Release Checklist — Version bump process