xiaotianfotos/homerail

What it solves

HomeRail transforms one-off AI agent chats into structured, auditable, and reusable workflows. It addresses the "black box" nature of chat sessions by organizing agent work into a Directed Acyclic Graph (DAG), making the process inspectable, replayable, and easier to improve. It is specifically designed for home servers or NAS environments to minimize the human attention required for automation.

How it works

HomeRail uses a DAG engine to orchestrate multiple agents with explicit handoffs. The system is structured as an inverted funnel:

• Voice Surface: Acts as the primary input method to reduce user friction, using ASR/TTS/VAD to collect intent.
• Generative UI: Instead of raw logs, the system produces structured, easy-to-read views of the agent's progress.
• DAG Runtime: The core engine that manages multi-agent orchestration. It uses Docker to provision isolated Worker containers for each node in the DAG, ensuring workspace isolation per run.
• Smart Brain/Efficient Workers: It allows different models to be assigned to different nodes (e.g., a powerful model for planning and a cheaper one for execution) to optimize cost and context window usage.

Who it’s for

It is designed for users with a home lab, NAS, or home server who want a resident home-datacenter agent capable of handling complex automation via voice and a generated interface.

Highlights

• Auditable Workflows: Every handoff is traced, and every run is replayable via the hr CLI.
• Docker-backed Isolation: Each DAG node runs in its own container, sharing a workspace for the specific run.
• Multi-Model Orchestration: Supports mapping different providers and models to specific agents within a single workflow.
• Voice-First Interface: Built-in support for voice interaction to minimize user bandwidth.
• Agent-Driven CLI: The hr CLI is designed to be operated by other coding agents (like Claude Code or Codex) for rapid DAG development and debugging.