DataHub

DataHub is a pioneering open AI Workflow infrastructure developed by Fudan university’s Institute of Computing and Intelligent innovation and the Shanghai Key Laboratory of data science. Unlike conventional tool directories or static repositories, DataHub functions as a living, dual-engine ecosystem—integrating a massive dataset discovery platform (510,000+ curated datasets) with a globally leading AI agent Skill hub (1.46M+ skills). Its mission is to make data and capabilities ‘discoverable, underStandable, comparable, and callable’—enabling seamless transitions from research to real-world execution.

What it does

DataHub bridges critical gaps in the AI Agent lifecycle. It aggregates, evaluates, and delivers AI Skills—modular, executable functions that empower Agents to perfORM tasks like coding, document summarization, Cloud Infrastructure management, or social media automation. Crucially, it goes beyond listing: every skill undergoes automated, multi-dimensional assessment—including quality, safety risk, and computational cost—generating a transparent ‘health report’. Simultaneously, its dataset engine links data to papers, methods, and corresponding skills, forming a unified knowledge graph for end-to-end AI Development.

Key capabilities

• Intelligent Skill Discovery: Combines keyword search, tag filtering, conversational recommendation, and scenario-based skill bundles—enabling users to find, compare, and deploy skills in SEConds.
• Full-Cycle Skill evaluation: Uses genericAgent—a research-grade AI agent—to audit each skill across six dimensions and 30+ metrics, covering functionality, security exposure, token usage, latency, and environmental dependencies.
• Agent-First Integration: Offers standardized APIs and documentation for seamless skill loading, chaining, and autonomous retrieval by AI Agents—including openclaw—supporting true self-evolution through on-the-fly skill composition.
• Data–Skill Fusion: Cross-links datasets (e.g., image classification benchmarks) with related training, inference, and Testing skills, enabling full-stack Workflows without context switching.

Who it is for

DataHub serves researchers replicating SOTA models, developers building production AI agents, Enterprise teams deploying secure Automation, educators teaching agent design, and students exploring hands-on AI Engineering. Its academic foundation ensures rigor and openness, while its scale and inteRoPErability meet industry-grade demands.

How to get started

Access is entirely web-based and free. Users can begin immediately at the public portal to browse skills by category (Coding, Productivity, AI-ML, Infra), filter by language or safety rating, or ask natural-language questions like ‘I need a Python script to clean CSV files’. No registration is required for exploration; contributors may subMIT skills via open channels. Integration with agents requires API keys and SDKs documented on the site.

Pricing and access caveat

DataHub is fully open and Free to use, with no credits, subscriptions, or paywalls. As a univeRSIty-led initiative, it prioritizes accessibility over Monetization. However, advanced features—such as private skill registries or enterprise-grade SLAs—may vary in future releases. The platform currently operates under a Chinese domain and may require regional network access for optimal performance outsIDE mainland China.