#python

GLM-OCR is a multimodal OCR model built on the GLM-V encoder-decoder architecture, specifically designed for complex document understanding. It integrates an advanced CogViT visual encoder and a GLM-0.5B language decoder, enhanced by Multi-Token Prediction (MTP) loss and stable full-task reinforcement learning to significantly improve training efficiency, recognition accuracy, and generalization. Achieving state-of-the-art performance with a 94.62 score on OmniDocBench V1.5, GLM-OCR excels in formula, table recognition, and information extraction, optimized for challenging real-world layouts. With only 0.9B parameters, it supports efficient deployment via vLLM, SGLang, and Ollama, offering low inference latency and reduced compute costs, making it ideal for high-concurrency services and edge deployments. Fully open-sourced with a comprehensive SDK, GLM-OCR is easy to integrate, providing robust and high-quality OCR for diverse practical business scenarios.

Granian is a high-performance HTTP server built in Rust, leveraging Hyper and Tokio, specifically designed for Python applications. It offers a unified and correct HTTP/1 and HTTP/2 (with planned HTTP/3 support) implementation, fully supporting ASGI/3, RSGI, and WSGI application interfaces. This product aims to streamline deployment by avoiding complex dependency stacks, providing superior concurrency and stable high-throughput performance, particularly excelling in WebSocket connections. Granian also supports HTTPS, mTLS, and direct static file serving, making it an ideal choice for modern, single-dependency Python service deployments prioritizing high performance and concurrency.

分享 GitHub 上有趣、入门级的开源项目。Share interesting, entry-level open source projects on GitHub.

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.

Agent Lightning, developed by Microsoft, is a versatile and framework-agnostic training platform designed to optimize AI agents with minimal code changes. It empowers developers to enhance the performance of agents built with any framework, such as LangChain, AutoGen, or even custom Python implementations, by leveraging advanced algorithms like Reinforcement Learning, Automatic Prompt Optimization, and Supervised Fine-tuning. A core strength is its ability to selectively optimize individual agents within complex multi-agent systems. This lightweight platform provides a clear pathway from initial agent deployment to steady, algorithm-driven refinement, fostering robust and adaptive AI agent development.

supervision is an open-source computer vision utility library developed by Roboflow, dedicated to providing an efficient and reusable toolset to streamline the development and deployment of computer vision applications. This product features model agnosticism, allowing seamless integration with leading classification, detection, and segmentation models such as Ultralytics and MMDetection. It offers a wide range of customizable annotators for clear and efficient visualization of model outputs, alongside robust dataset management capabilities that support loading, splitting, merging, and converting data in various formats like COCO, YOLO, and Pascal VOC. Supervision significantly enhances developer efficiency in data handling, model inference, and result presentation, making it particularly suitable for real-time video stream analysis, object tracking, and behavioral analysis scenarios, including dwell time analysis and vehicle speed estimation.

Deep-Live-Cam is an efficient AI Agent product for real-time face swap and video deepfake. It enables one-click real-time face swapping and video deepfaking with just a single source image, significantly streamlining the operation process. Key features include real-time Mouth Mask for retaining natural dialogue movements and Face Mapping capabilities for simultaneously applying different faces to multiple subjects. Deep-Live-Cam is widely applied in AI-generated media creation, live performances, and entertainment content production (e.g., movie experiences, social interactions, meme creation). It supports multi-platform operation on NVIDIA, AMD GPUs, CPUs, and Mac Silicon, offering convenient pre-built versions that balance ease of use with performance. Furthermore, the product incorporates robust built-in content moderation mechanisms, emphasizing responsible, legal, and ethical usage.

AutoAgent by HKUDS is a cutting-edge, fully-automated, and zero-code framework for Large Language Model (LLM) agents. It empowers users to effortlessly build, deploy, and orchestrate complex LLM agent systems purely through natural language, without requiring any coding expertise. Its core strength lies in intelligent agent construction, tool creation, and workflow generation capabilities, dynamically optimizing and adapting workflows based on high-level task descriptions. The framework includes a robust “user mode” that functions as an AI research assistant, excelling in information retrieval, complex analytical tasks, and comprehensive report generation. Additionally, its “agent editor” and “workflow editor” modes enable personalized tool and agent customization via conversational interactions, significantly lowering the barrier to AI development and accelerating the implementation of intelligent applications across industries.

RedditVideoMakerBot is an innovative AI agent designed to fully automate the creation of engaging short-form videos from Reddit content, eliminating the need for manual video editing or asset compilation. With a single command, users can generate viral-potential videos, streamlining the process of gathering and editing materials for platforms like TikTok, YouTube, and Instagram. Key features include customizable background music, selection of specific Reddit threads or subreddits, diverse video backgrounds, voice customization, and filters for NSFW content, all managed via a `config.toml` file and Reddit API integration. The bot outputs a ready-to-upload video file, empowering creators to produce content efficiently and at scale.

Mem0 offers a universal intelligent memory layer designed for AI Agents and assistants, enabling personalization and continuous learning. It seamlessly retains context and user preferences through multi-level memory (User, Session, and Agent states). Leveraging an innovative V3 memory algorithm that integrates single-pass extraction, entity linking, and multi-signal retrieval, Mem0 significantly enhances memory accuracy and recall. With developer-friendly APIs and cross-platform SDKs supporting various LLMs and embedding models, Mem0 is widely applicable in scenarios like intelligent customer support, personalized assistants, and healthcare, empowering the creation of AI systems with deeper user understanding.

tradingview-mcp is an advanced AI-powered trading intelligence framework and MCP server designed for real-time market analysis and informed trading decisions. It integrates real-time crypto and stock screening, over 30 professional technical indicators (e.g., Bollinger Bands, RSI, MACD), a multi-strategy backtesting engine, and live social media sentiment analysis. The framework deploys specialized AI agents (Technical Analyst, Sentiment & Momentum Analyst, Risk Manager) that collaboratively analyze data to provide high-confidence buy/sell signals. Its open-source nature, rapid deployment, and broad compatibility with Claude Desktop and OpenClaw (supporting Telegram, WhatsApp, etc.) make it a powerful, cost-effective trading intelligence solution for individual investors and quantitative analysts.

AI-Trader, developed by HKUDS, is a 100% fully-automated, agent-native trading platform specifically designed to provide AI agents with a dedicated environment for financial market interaction. It enables rapid integration for leading AI agents like OpenClaw and nanobot via a single command, fostering collective intelligence trading. Key features include cross-platform signal synchronization, one-click copy trading, universal market access (stocks, crypto, forex, etc.), and a reward system for signal publication and followers. AI-Trader empowers agents to exchange strategies and execute trades, while also offering human traders risk-free paper trading and copy trading options, aiming to build an AI-driven, collaborative financial trading ecosystem that enhances market efficiency and intelligence.

Hermes Agent, developed by Nous Research, is a self-improving AI agent distinguished by its unique built-in learning loop. It autonomously creates and refines skills from experience, intelligently persists knowledge across sessions, and continuously builds a deepening user model. Supporting a wide array of LLMs from various providers without vendor lock-in, it offers exceptional model flexibility. Hermes Agent provides a robust terminal interface and seamless cross-platform interaction via Telegram, Discord, and other messaging services. Key features include scheduled automations, delegation to isolated subagents, and flexible deployment options ranging from cost-effective VPS to serverless infrastructure, making it a versatile, adaptive, and research-ready AI companion.

AI Hedge Fund is an AI-powered multi-agent proof-of-concept project designed to explore the application of artificial intelligence in simulated trading decisions. This modular multi-agent system integrates 13 specialized investor agents, embodying diverse philosophies from renowned figures like Buffett, Graham, Wood, Ackman, and Taleb. These are complemented by dedicated agents for valuation, sentiment analysis, fundamental and technical analysis, risk management, and portfolio management. The system generates trading signals, calculates intrinsic stock values, analyzes market data, performs risk assessments, and makes simulated portfolio decisions. Developed solely for educational and research purposes, it offers both command-line and web application interfaces, supports various large language models, and includes comprehensive backtesting capabilities, strictly without engaging in actual trading.