Loushang is an AI-native agent harness designed to address breakdowns in complex AI workflows, such as lost context, unresumable execution, and poor tool governance. It provides a method-native AI work system for running complex tasks from intent to verified delivery, targeting engineers, researchers, and power users. The primary benefit is making complex AI-driven work more reliable, recoverable, auditable, and deliverable, particularly within software development contexts.

How It Works

Loushang treats core components like methods (structured work contracts defining roles, phases, and expectations), sessions (durable coding conversations with state persistence), tools (governed executable capabilities), and extensions (project-level code hooks) as first-class runtime objects. This approach aims to enhance the reliability, recoverability, and auditability of complex AI agent execution by managing context, state, and tool usage systematically, moving beyond simply improving agent intelligence.

Quick Start & Requirements

The recommended installation is from source. Clone the repository, navigate to the directory, and set up a development environment using uv:

git clone https://github.com/zhnt/loushang.git
cd loushang
uv venv .venv
source .venv/bin/activate
uv pip install -e ".[dev]"

Alternatively, make bootstrap creates the .venv/ environment and installs the project in editable mode. Basic commands like loushang --help and loushang --list-models can then be run. Prerequisites include Git and Python, managed via uv.

Highlighted Details

• Supports multi-model LLM orchestration, routing across providers like GPT, Claude, DeepSeek, Qwen, Kimi, GLM, and MiniMax.
• Features persistent coding sessions with capabilities for resuming, forking, exporting, and diagnostics.
• Provides built-in coding tools and configurable tool surfaces with governance policies.
• Includes an AI SDK (loushang.ai) offering model registry, streaming, tool calls, and cost helpers.
• Enables method-guided coding and reusable workflow assets through extensions.

Maintenance & Community

Loushang was initiated by Heng Zhou. For inquiries, feedback, collaboration, or community group invitations, contact zhnt@foxmail.com. The project is in active early development, with loushang code and the loushang.ai SDK as the current stable focus.

Licensing & Compatibility

The project is licensed under the Apache License 2.0. Redistribution requires retaining the LICENSE and NOTICE files and including attribution in product documentation. This license is generally permissive for commercial use and integration with closed-source projects.

Limitations & Caveats

Loushang is in early development, and broader work surfaces like loushang work, loushang research, and loushang ppt are considered evolving roadmap items. The Makefile currently supports local development (make bootstrap) and local binary installation (make install-binary) but does not offer a standard make install target for broader deployment.