openJiuwen-ai/agent-core

What it solves

openJiuwen Core provides a high-performance runtime and Python SDK for building large language model (LLM) applications. It addresses the difficulty of creating stable, production-ready agents by offering a balance between flexible reasoning and structured workflow orchestration, while solving common performance bottlenecks like slow execution and the lack of state recovery for interrupted tasks.

How it works

The framework operates as a core engine that separates the SDK interface layer from the agent engine. It allows developers to build agents using two primary paradigms:

• ReActAgent: Uses a "thinking $\rightarrow$ action $\rightarrow$ observation" loop for dynamic reasoning and self-correction.
• WorkflowAgent: Executes predefined, multi-step processes for structured task automation.

It utilizes an asynchronous parallel graph executor to handle component concurrency and streaming IO, and includes a resource manager to handle agent state saving and recovery (checkpoints).

Who it’s for

It is designed for developers building LLM-powered applications who need a scalable, high-performance runtime that supports both autonomous reasoning and strict process automation.

Highlights

• Dual Agent Paradigms: Supports both ReAct-style autonomous agents and process-driven WorkflowAgents.
• High-Performance Engine: Features an asynchronous parallel graph executor for concurrent component execution and streaming processing.
• State Management: Enables agent state saving and interruption recovery, allowing users to switch between different workflows in a single session.
• Prompt Tooling: Includes built-in capabilities for one-click prompt generation and automatic optimization using real-world datasets.
• Observability: Provides full-link observability and debugging tools for agent optimization.