Agent Tools Repository

This is a code repository for building and managing intelligent agent tools, containing multiple projects and modules related to AI agents.

Top-Level Directory Structure

Directory	Description
packages/	Python package directory, containing all installable sub-packages
packages/openjiuwentools_core/	Core package, providing foundational functionality
packages/openjiuwentools_infer_router/	Inference routing package, providing a KV Cache-aware routing system
openJiuwen-vllm-affinity/	Integration project between OpenJiuwen and vLLM, supporting affinity scheduling
Agent Innovation Challenge/	Code and resources related to the Agent Innovation Challenge

Package Management Mechanism

This project adopts a modular package management mechanism, where each sub-package is independently built as a wheel package:

Source directory	Wheel package name
packages/openjiuwentools_core	openjiuwentools-core-{version}-py3-none-any.whl
packages/openjiuwentools_infer_router	openjiuwentools-infer-router-{version}-py3-none-any.whl

Installation

Option 1: Install the meta-package (recommended)

# Install the core package (minimal dependencies)
pip install openjiuwentools

# Install a sub-package for a specific feature
pip install "openjiuwentools[infer-router]"

# Install all commonly used sub-packages
pip install "openjiuwentools[top]"

Option 2: Install sub-packages directly

# Install a specific sub-package individually
pip install openjiuwentools-infer-router
pip install openjiuwentools-core

Option 3: Development mode install (from source)

git clone https://gitcode.com/openJiuwen/agent-tools.git
cd agent-tools

# Install with uv (recommended)
uv sync --extra top

# Or install in editable mode with pip
pip install -e ".[top]"

Development Environment Setup

1. System Requirements

• Python 3.10 or higher
• Git
• Linux or macOS recommended

2. Clone the Repository

git clone https://gitcode.com/openJiuwen/agent-tools.git
cd agent-tools

3. Install Dependency Management Tools

It is recommended to manage dependencies with pip or uv:

# Install uv with pip (optional)
pip install uv

4. Virtual Environment Setup

Create a unified virtual environment for the entire repository:

# Create a virtual environment in the repository root
python3 -m venv .venv
source .venv/bin/activate  # Linux/Mac
# or
.venv\Scripts\activate  # Windows

# Install project dependencies
pip install -e ".[top]"
# or with uv
uv sync --extra top

pre-commit Configuration

1. Install pre-commit

# Global install (recommended)
pip install pre-commit

# Or install within a virtual environment
pip install -e ".[dev]"

2. Install Hooks

Run in the repository root:

pre-commit install

3. pre-commit Configuration Notes

The repository uses the following pre-commit hooks:

• ruff: fast Python code linting and formatting tool
• codespell: source code spell checker
• typos: another source code spell checker
• markdownlint-cli: Markdown file format checker

Hook configuration is in the .pre-commit-config.yaml file at the repository root.

4. Run pre-commit Manually

# Check all files
pre-commit run --all-files

# Check specific files
pre-commit run --files file1.py file2.md

Development Notes

1. Code Style

• Follow the PEP 8 code style guide
• Use ruff for code formatting and linting
• Keep code concise and clear, with appropriate comments

2. Commit Conventions

• Ensure code passes all pre-commit checks before committing
• Commit messages should be clear and accurate, describing the specific changes
• Avoid including unrelated changes in the same commit

3. Branch Management

• Use main as the stable branch
• Use feature/feature-name branches for feature development
• Use fix/bug-name branches for bug fixes

4. Testing

• Write unit tests for new features

• Ensure all tests pass before committing code

• Run tests with pytest:

  pytest

5. Documentation

• Update relevant documentation to reflect code changes
• Keep API documentation accurate and complete
• Write documentation in Markdown format

Project-Specific Notes

openjiuwentools-infer-router package

• Provides LLM inference routing and scheduling
• Supports multiple load balancing algorithms and scheduling strategies
• Includes a complete API service layer and monitoring system
• KV Cache-aware routing, supporting vLLM and SGLang

openJiuwen-vllm-affinity project

• Integrates OpenJiuwen with vLLM
• Supports affinity-based scheduling strategies
• Optimizes large model inference performance

Contact

If you have questions or suggestions, please reach out via:

• Project homepage: https://gitcode.com/openJiuwen/agent-tools
• Issue tracker: https://gitcode.com/openJiuwen/agent-tools/issues
• Discussions: https://gitcode.com/openJiuwen/agent-tools/discussions