Runnable, documented projects worth studying and extending.
AI and AgentsOpenViking - AI Agent Context DatabasePWByteDance's open-source AI Agent context database manages the memory, resources, and skills required by agents through a file system paradigm, supporting layered context transmission and self-evolution.
OthersMiroFishPWA swarm intelligence prediction engine based on multi-agent technology builds a high-fidelity digital world for social simulation and future simulation, supporting scenarios such as public opinion analysis and financial forecasting
OthersSupervisionPWComputer vision toolkit simplifies visualization and analysis of object detection, segmentation, and tracking
Efficiency toolsBolt.newPWStackBlitz's full-stack AI development tool builds complete applications through conversation within the browser
AI and AgentsCrewAIPWA multi-agent collaboration framework allows AI agents to work together like a team, dividing tasks to complete complex tasks
OtherspromptfooPWLLM application evaluation and red team testing tools. Supports automated prompt evaluation, model comparison, and security vulnerability scanning, and can integrate CI/CD pipelines.
OthersBettaFish WeiyuPWA multi-agent public opinion analysis system implemented from scratch, covering 30+ domestic and international social media platforms, automatically collecting and analyzing millions of comment data. Supports AI-driven full-domain monitoring, multimodal analysis, and agent forum collaboration mechanisms.
OthersLangServePWDeploy LangChain chains and agents as one-click tools for REST APIs
Efficiency toolsAn AI tool for long-video meeting minutes, interview program editing, and course overview: whisperVideoPWwhisperVideo is an AI tool that can convert speech in long videos into text, automatically recognize the speaker, match the content to the face in the frame, and generate subtitles and visual panels with the speaker's ID. It supports hour-level footage, automatically dividing scenes and paragraphs, suitable for interviews, film editing, and meeting minutes reviews. Using WhisperX for text transcription, Pyannote for speaker separation, combined with SAM3 for face detection, and finally integrating all information to generate subtitles and panel views
OthersHermes AgentPWAn AI Agent produced by NousResearch, an agent that grows alongside you, supports adaptive learning and skill expansion, and has a 5k+ star rating.
Othersnotebooklm-pyPWGoogle NotebookLM is an unofficial Python API that supports programmatic access to all NotebookLM features (including Web UI capabilities not exposed), and can be integrated with AI agents such as Claude Code and Codex.
OthersMiroFishPWA simple and universal swarm intelligence engine that can predict everything. Based on swarm intelligence algorithms, supports multi-domain predictive analytics, with a star rating of 16k+.
Efficiency toolsvLLMPWA high-throughput LLM inference and service engine using PagedAttention technology, 24 times faster than HuggingFace
OthersGemma 2PWGoogle's open-source lightweight large model, available in 2B and 9B versions, suitable for local deployment
Efficiency toolsVery cool—Tencent has just open-sourced its latest 3D generation model: Hunyuan 3D version 2.0, which produces more detailed geometric structures, richer texture colors, and performance surpassing existing open-source and commercial modelsPWTencent has open-sourced version 2.0 of Hunyuan 3D, featuring finer geometric structures and rich texture colors, supporting high-resolution 3D asset generation, including geometric generation and texture synthesis models, providing a convenient production platform called Hunyuan3D-Studio.
Efficiency toolsStagehandPWBrowserbase is an open-source AI web automation SDK that controls browsers using natural language
Efficiency toolsllama.cppPWAn LLM inference engine implemented purely in C/C++, supporting CPU and various hardware components, lightweight and efficient
OtherstestPWtest
OthersWebnovel WriterPWBased on Claude Code, this long-form online novel writing support system uses RAG+ entity graphs to solve AI writing forgetting and hallucination issues, supports 2 million word serializations, and features a built-in tracking system and visualization dashboard.
Efficiency toolsFlowisePWDrag-and-drop LLM application building tools visualize LangChain workflows, enabling zero-code AI workflow construction
Efficiency toolsTencent has also open-sourced a universal 3D reconstruction model: HunyuanWorld-Mirror, which adds diverse inputs and can generate 3D worlds with one click for videos and multi-angle imagesPWTencent has open-sourced the HunyuanWorld-Mirror model, supporting high-quality 3D world generation from videos and multi-perspective images. Using a feedforward architecture, it can simultaneously generate multiple 3D representations, suitable for e-commerce, museum artifact digitization, and other scenarios
OthersHermes AgentPWNousResearch's adaptive AI agent framework supports tool calls, memory management, and continuous learning, claiming to be an agent that grows alongside you
Efficiency toolsHKUDS has launched a hugely popular AI personalized learning tool: DeepTutor. It solves three problems: too much information to find answers, no one to explain or understand complex knowledge, and no one to set questionsPWDeepTutor is an AI personalized learning tool that can solve problems such as difficulty in searching for information, insufficient understanding of complex knowledge, and lack of practice questions. Through interactive visual learning, it breaks down complex concepts into intuitive diagrams and generates exercises or mock papers based on users' knowledge levels and learning goals, greatly improving learning efficiency
Efficiency toolsPerplexicaPWAn open-source AI search engine, an alternative to Perplexity, supporting multiple search modes