loushang
AI-native agent harness for coding workflows by python: multi-model LLM orchestration, stateful sessions, tool governance, traceable delivery, and provider routing for GPT, Claude, DeepSeek, Qwen, Kimi, GLM, and MiniMax.
Loushang is a layered AI-era operating system framework built in Python, providing a kernel, protocol layer, adapters, and extension points for building agent-driven workflows in a monorepo.
Loushang is described as an operating system for the AI era, aimed at helping individuals, teams, and organizations find opportunities, manage complexity, and move quickly. The README is written in Chinese. Beyond that framing, the description is brief.
The project follows a layered architecture with four parts: a kernel that defines how the system runs, a protocol layer that defines how it communicates with the outside world, adapters that connect it to different environments and interfaces, and extension points that allow customization without breaking consistency. The code is organized as a monorepo containing several packages: loushang-ai, loushang-agent, loushang-channel, loushang-tui, loushang-methods, and loushang-coding. The package names suggest coverage of AI integration, agent runtime, a terminal UI, and a coding-focused component. The README links to internal architecture documents covering topics like AI streaming, agent types, and a channel boundary protocol, but those documents are not reproduced in the README itself.
Where it fits
- Build a custom AI agent runtime that routes tasks through different LLMs using the adapter layer.
- Create a terminal-based interface for monitoring and managing AI workflows with loushang-tui.
- Extend the framework with a coding assistant component via loushang-coding without breaking the core architecture.