Runnable, documented projects worth studying and extending.
OthersSemantic RouterPWThe vLLM project launched a system-level intelligent router supporting multi-model hybrid routing in cloud, data center, and edge scenarios
OthersPocketPal AIPWAn open-source app that runs large language models directly on mobile phones, supporting multiple model formats and allowing users to experience AI conversations on mobile devices without servers
OtherssemPWSemantic-level version control CLI - Tree-sitter-based support for entity-level diff, blame, dependency graph, and impact analysis in 21 languages
OthersArchonPWThe first open-source AI coding Harness builder enables repeatable and deterministic execution of the AI coding process. For example, Dockerfile is for infrastructure, GitHub Actions is for CI/CD, and Archon is for AI coding workflows. Supports YAML definition of development processes.
OthersOpenDataLoader PDFPWAI-ready PDF parser, ranked first in benchmark (0.907), supporting complex tables, formulas, and charts. Local deterministic mode + AI hybrid mode, multilingual SDK for Python/Node.js/Java.
OthersRustFS - Rust high-performance object storagePWA high-performance distributed object storage system built in Rust, 2.3 times faster than MinIO, fully compatible with the S3 protocol, suitable for data lakes, AI, and big data scenarios
OthersAirLLMPWRunning 70B large model inference on a single 4GB VRAM GPU! Through layered loading and quantization technology, consumer-grade hardware can run ultra-large models, achieving 15.6K stars.
OthersPlannotatorPWVisualize annotation and review of AI coding agent plans and code diffs, support team sharing, and send feedback to agents with one click. This allows humans to better supervise and guide the AI coding process.
OthersVoiceBoxPWAn open-source speech synthesis studio supporting high-quality TTS, tone cloning, and style transfer. 18k stars, 3k+ growth this week, one of the hottest projects in the AI voice field.
AI and AgentsJCode - Ultra-high-performance encoding agent frameworkPWJCode is a next-generation coding agent framework designed for multi-session workflows and extreme performance. Built-in semantic vector memory system, boot speed 245 times faster than Claude Code, with 10 sessions occupying only 117MB of memory (Claude Code occupying 2.3GB).
OthersMaigretPWOSINT open-source intelligence tool automatically collects personal digital footprint profiles on 3000+ websites via username, supporting report generation and data export
OthersGoogle AI Edge GalleryPWGoogle's on-device ML/GenAI application showcase supports local experience and use of AI models on devices, with rich examples and use cases
OthersLiteRT-LMPWGoogle AI Edge launched an on-device large model inference engine, written in C++, optimized for mobile devices and embedded platforms, supporting efficient local inference for models like Gemma/Phi.
OthersPersonaPlexPWNVIDIA's open-source multi-character character generation framework can batch create AI characters with consistent styles, supporting fine control over dialogue style, personality traits, and background stories.
Efficiency toolsOpenScreen - Free open-source screen recording toolPWFree open-source screen recording and presentation production tool, watermark-free, no subscription, commercially usable. The open-source alternative to Screen Studio, 23k+ stars, added 2700+ stars today.
OthersGitNexusPWBrowser-side zero-server code intelligent engine; just drag into GitHub repository or ZIP files to generate interactive knowledge graphs, with built-in Graph RAG Agent, suitable for code exploration and understanding
OthersClaude Skills CollectionPW66 Claude Code professional skill packs designed specifically for full-stack developers. Make Claude Code your expert-level paired programming partner, covering frontend, backend, DevOps, security, and more.
OthersShannonPWKeygraphHQ's open-source proprietary white-box AI penetration testing tool analyzes source code to identify attack vectors and performs real vulnerability verification. 36k+ Star, designed specifically for web applications and API security testing
OthersqmdPWLocal document search engine CLI — supports semantic search of knowledge bases, meeting notes, and other documents, tracks the latest SOTA methods, runs fully locally, prioritizing privacy. Developed by Shopify CEO Tobi Lütke.
OthersHonchoPWAI Agent Memory Library: Replacing complex RAG pipelines with a serverless architecture, providing agents with instant retrieval and long-term memory capabilities. Supports stateful agent construction.
OthersGoose (aaif)PWOpen-source, scalable AI Agent: not only code suggestions but also installation, execution, editing, and testing, supporting any LLM. Block's Goose community enhancement version, 38k stars.
OthersSub2APIPWAI API gateway platform supports multi-model subscription quota distribution and management for Claude, OpenAI, Gemini, and others, one-stop transfer services, and supports ride-sharing services, intelligent scheduling, and precise billing.
OthersMeridianPWAllows Claude Max subscribers to use Claude's proxy tools on third-party tools (OpenCode, Aider, Cline, etc.). Bridges with Anthropic's official SDK to address API usage pain points for subscribers.
Efficiency toolsAndrej Karpathy Claude Code skill documentPWBased on Andrej Karpathy's observations on LLM programming pitfalls, distilled into a CLAUDE.md file containing four main principles: think before coding, prioritize simplicity, make precise modifications, and execute goal-driven. Help Claude Code avoid over-engineering and misconceptions.