Homebot: A Personal AI Agent for Conversational Home Assistance and Automation

Homebot is a locally hosted AI agent designed to bridge the gap between text-based LLM workflows and the specific requirements of household environments. Unlike standard chatbot interfaces, Homebot is built to handle the social and physical constraints of a home, where users often need hands-free interaction while multitasking. By decoupling the request-processing logic from the specific communication channel (voice, Telegram, or Feishu), the system provides a consistent interface for home automation and assistance.

The core of Homebot is a layered architecture consisting of channels, a message bus, an agent runtime, and a tool/skill registry. When a user sends a request, the channel normalizes the input into a standard format, which is then passed to the Agent Runtime. This runtime manages the interaction loop: it constructs the context, invokes the LLM, and executes registered tools or retrieves task-specific skills as needed.

Crucially, Homebot separates session management from the execution path. Messaging history is scoped to specific chat identifiers, ensuring that conversations remain isolated by channel. For voice, the system uses a wake-word-triggered, bounded session that automatically cleans up after interaction, preventing stale context from bleeding into future requests.

To support natural, hands-free conversation, Homebot implements a structured dialogue-state protocol. Instead of relying on the model to guess when a conversation ends, the system uses an explicit state machine (STOPPED, LISTENING, RECOGNIZING, THINKING, PLAYING). The model provides a dialogue state—end, follow_up, or continuous—which dictates how the channel handles the transition after the response is delivered. This allows the system to seamlessly switch between one-off commands and multi-turn discussions without premature termination.