OpenSpace provides a self-evolving engine designed to enhance AI agents, making them smarter, more cost-efficient, and capable of continuous learning and adaptation. It addresses the critical weakness of current agents: their inability to learn from real-world experience, leading to wasted tokens and repeated failures. By enabling agents to evolve their skills and share knowledge collectively, OpenSpace aims to significantly reduce operational costs and improve performance across a wide range of professional tasks. The target audience includes developers, researchers, and power users working with various AI agent frameworks.

How It Works

OpenSpace functions as a "self-evolving engine" that integrates with existing AI agents. Its core approach revolves around three key superpowers: Self-Evolution Skills, Collective Agent Intelligence, and Token Efficiency. The Self-Evolution engine automatically repairs broken skills (AUTO-FIX), improves successful patterns (AUTO-IMPROVE), and captures winning workflows from actual usage (AUTO-LEARN), all while continuously monitoring skill quality. Collective Agent Intelligence transforms individual agents into a shared brain, where one agent's improvement becomes an upgrade for all connected agents, creating network effects. This continuous learning and sharing mechanism dramatically reduces token consumption by reusing successful solutions instead of reasoning from scratch, claiming up to 4.2x better performance with 46% fewer tokens.

Quick Start & Requirements

Installation involves cloning the repository (git clone https://github.com/HKUDS/OpenSpace.git) and installing dependencies (pip install -e .). A lightweight clone option is available to skip the assets/ folder. The project requires Python 3.12+ and Node.js ≥ 20 for the local dashboard. Community exploration and skill browsing are available without installation at open-space.cloud.

Highlighted Details

• Performance Benchmark: On the GDPVal benchmark (50 real-world professional tasks), OpenSpace agents earned 4.2x more income than baseline agents using the same backbone LLM (Qwen 3.5-Plus), while capturing 72.8% of task value.
• Cost Efficiency: Achieved 45.9% reduction in token usage between Phase 1 (cold start) and Phase 2 (warm rerun) on the GDPVal benchmark, demonstrating significant cost savings as skills evolve.
• Autonomous Development: Showcased by "My Daily Monitor," a complex system built entirely by an agent using over 60 evolved skills from scratch, demonstrating end-to-end autonomous software development capabilities.
• Skill Focus: Analysis reveals that most evolved skills prioritize tool reliability and error recovery (e.g., file I/O, execution recovery) over task-specific knowledge, forming a robust foundation for agent operation.

Maintenance & Community

OpenSpace fosters community through Feishu and WeChat groups, with links provided in COMMUNICATION.md. The project has a public roadmap outlining future developments such as Kanban-style orchestration and collaboration pattern evolution. Community skills and evolution lineage can be explored at open-space.cloud.

Licensing & Compatibility

The project is released under the MIT License, which generally permits broad use, including commercial applications, without significant copyleft restrictions.

Limitations & Caveats

The README does not explicitly detail limitations such as alpha status or known bugs. However, the project's extensive feature set and ongoing roadmap suggest active development. The initial repository clone size can be substantial due to the assets/ folder, though a workaround is provided.