Launch Multi-Agent AI Systems with OxyGent in Just 20 Lines of Code

OxyGent enables developers to flexibly compose AI agents like building blocks, offering extreme scalability, full‑trace decision making, and top performance on the GAIA Benchmark.

Installation

Required environments:

Python 3.10+ (create a conda environment)

conda create -n oxy_env python==3.10

conda activate oxy_env

OxyGent package pip install oxygent Node.js is needed only when using MCP tools (download from the official site).

Hello World

Create a .env file with your large‑model credentials:

DEFAULT_LLM_API_KEY = "<your_key>"
DEFAULT_LLM_BASE_URL = "<your_url>"
DEFAULT_LLM_MODEL_NAME = "<model_name>"

Run the starter command (see image).

RAG

MoA

Tool Integration

Local MCP

SSE MCP

FunctionHub

All registration methods (FunctionHub, LocalMCP, SSE‑MCP) produce the same runtime effect (see image).

External MCP

Auto Top‑K Tool Retrieval

Block‑Style Multi‑Agent Construction

Hierarchical Agents

Workflow Integration

Reflexion Mechanism

Rapid Deployment

Data Persistence

The framework provides a robust storage mechanism for later SFT or RL training.

Concurrency Limits

Multi‑Environment Configuration

Distributed Execution

Advanced Features

Multimodal support

Weight‑based memory filtering during execution

Retrieval of additional tools

Custom large‑model output parsers

Custom SSE interfaces

Result post‑processing or formatting

Agents invoking multiple tools simultaneously

Restart tasks from intermediate nodes

Plan‑and‑Solve paradigm

Open‑Source Invitation

Join the OxyGent community on GitHub (https://github.com/jd‑opensource/OxyGent) and contribute to the project’s evolution.