#llm

AI-Search-Hub is an open-source AI Skill designed to aggregate native search and web extraction capabilities from leading AI platforms such as Gemini, Grok, Doubao, Yuanbao, and Kimi, establishing an efficient, reusable information hub. It addresses the challenges of traditional data acquisition, including crawler maintenance, platform risk control, and data cleaning. Users can leverage a "one query, all search skill" approach to freely obtain diverse information like tech trends, industry sentiment, hot topics, and travel plans. Through this hub, Agents and workflows can harness the powerful understanding and processing capabilities of major AI platforms to efficiently access hard-to-reach data sources like WeChat official accounts, Douyin, and X/Twitter. It automates data collection, cleaning, and organization, delivering unified, structured results to significantly enhance information retrieval efficiency and quality.

Rowboat is an open-source, local-first AI coworker designed to build and leverage a long-lived knowledge graph from your digital workspace. By connecting to services like Gmail and Google Calendar, it transforms your emails and meeting notes into an inspectable, Obsidian-compatible Markdown vault stored locally. This unique approach allows Rowboat to remember important context and help you act on it—from drafting documents and preparing meeting briefs to tracking topics and automating follow-ups. Emphasizing privacy and local data storage, Rowboat offers a compounding memory system that understands and anticipates your needs, supporting both local and hosted AI models, and extensible through external tools for enhanced productivity.

MemPalace is a leading open-source AI memory system designed to permanently store and efficiently retrieve every interaction users have with AI. It introduces a unique "Memory Palace" architecture, organizing raw conversations into hierarchical structures for high-fidelity recall and navigability. The system achieved the highest-ever LongMemEval score of 96.6% in raw mode. MemPalace runs entirely locally, is free, protects data privacy, supports multiple data mining modes, and integrates seamlessly with popular AI tools like Claude and ChatGPT, effectively solving the AI memory fragmentation problem.

Firecrawl is a robust Web Data API specifically engineered for AI Agents, empowering them with clean and high-quality web data. It facilitates scalable web search, scraping, and interaction, boasting industry-leading reliability with 96% web coverage, including JS-heavy pages, all without the complexities of proxy management. Delivering data at blazingly fast speeds with a P95 latency of 3.4 seconds, Firecrawl provides LLM-ready outputs such as clean Markdown, structured JSON, and screenshots, optimizing input for AI applications. The API autonomously handles proxy rotation, orchestration, rate limits, and JS-blocked content. It also supports media parsing (PDFs, DOCX) and enables AI agents to perform interactive actions like clicking, scrolling, and typing. Designed for seamless integration with any AI agent or MCP client, Firecrawl is an open-source solution.

LangGraph, developed by LangChain, Inc., is a low-level orchestration framework specifically engineered for building, managing, and deploying long-running, stateful multi-agent applications. By representing application logic as cyclical directed graphs (DAGs), it empowers developers with precise control over LLM execution flow, including when to halt, loop, or transfer control to other agents or humans. Key features include durable execution, seamless human-in-the-loop capabilities, comprehensive short-term and long-term memory management, and deep integration with LangSmith for robust debugging and production-ready deployment, making it an ideal choice for complex and reliable AI agent development.