All ClawAll Claw
Back to Catalog
OpenSpace logo
use-cases

OpenSpace

OpenSpace by HKUDS (University of Hong Kong) is an open-source self-evolving skill engine that empowers AI Agents to automatically learn, reuse, and optimize skills from every task, making automation smarter and significantly more token-efficient.

OpenSpace preview

Overview

OpenSpace is an open-source Self-Evolving Skill Engine developed by the HKU Data Science Lab (HKUDS) at the University of Hong Kong. It enables AI Agents to evolve continuously by capturing, deriving, and reusing skills from every task they perform, instead of reasoning from scratch every time.

The core idea is simple yet powerful: Agents should learn from experience like humans do. OpenSpace treats skills as living entities that can be automatically captured, fixed, improved, and even shared via the community cloud (open-space.cloud).

Key Features

  • Self-Evolving Modes: Supports FIX (auto-repair broken skills), DERIVED (create new skills from existing ones), and CAPTURED (extract new patterns from task execution).
  • Automatic Maintenance: Built-in health checks, auto-fixing, and performance optimization keep the skill library efficient over time.
  • Multi-Backend Support: Integrates with shell, GUI, MCP (Multi-Modal Control Protocol), web, system operations, and more — allowing Agents to actually control computers and external tools.
  • Significant Token Savings: Real benchmarks show up to 46% token reduction and 4.2× performance gains on GDPVal tasks.
  • Persistent Skill Database: Uses SQLite for long-term storage, manual editing, and multi-task pipelines.
  • Cloud Skill Sharing: Connect to https://open-space.cloud to share and discover community-evolved skills.
  • Easy Integration: Can run as a standalone Agent or integrate via MCP Server with frameworks like Claude Code, Cursor, etc.

Use Cases

  • Complex Task Automation: Let Agents handle software development, data analysis, web operations, and system management while continuously improving.
  • Long-Running Enterprise Agents: Build autonomous Agents that become more efficient and cheaper to run over time.
  • Multi-Agent Systems: Share evolved skills across multiple Agents for collaborative intelligence.
  • Cost Optimization: Dramatically lower LLM token costs in high-frequency automation scenarios.
  • AI Research: Study how Agents learn collectively and how skills propagate in real-world tasks.

Quick Start

  1. Clone the repository: git clone https://github.com/HKUDS/OpenSpace.git
  2. Set up environment variables (e.g., OPENAI_API_KEY) and configure config_agents.json / config_mcp.json.
  3. Run cold-start tasks — the Agent will automatically begin capturing skills.
  4. Repeat similar tasks to observe skill reuse, reduced token usage, and faster execution.
  5. (Optional) Connect to the cloud community for shared skills.

Detailed configuration examples and documentation are provided in the repo.

Technical Highlights

OpenSpace treats skills as dynamic, living assets rather than static prompts. It analyzes task execution traces to extract reusable patterns, supports multi-task pipelines, and enables continuous improvement. Benchmarks on 50+ real professional tasks demonstrate its practical value in building next-generation self-improving AI Agents.

OpenSpace is ideal for developers and researchers who want sustainable, cost-effective, and truly intelligent Agent automation.

GitHub: https://github.com/HKUDS/OpenSpace
Cloud Community: https://open-space.cloud

Tags

ai-agentopen-sourcehkuskill-engineself-evolvingautomationtoken-efficiencymulti-agentmcp