Open-source AI and machine learning tools on GitHub — LLM frameworks, model runtimes, agents and training libraries, ranked by stars and reviewed by TopGit.
Superpowers is an open-source (MIT) skills library and bootstrap instruction set that turns a coding agent's ad-hoc habits into a fixed pipeline: brainstorm a spec, write a plan, build under TDD, review, then close out the branch. Skills trigger automatically once installed, and the same methodology works across eleven different agent harnesses, each requiring its own install step.
TensorFlow is Google's open-source, end-to-end platform for machine learning, hosted at tensorflow/tensorflow under the Apache-2.0 license. It was originally built within Google Brain's Machine Intelligence team for ML and neural network research, and today it ships stable Python and C++ APIs alongside GPU, CPU-only, and Docker install paths. The README positions it as covering both research work and shipping ML-powered applications.
AutoGPT is an open-source platform designed for building, deploying, and running AI agents that can carry out complete workflows. Users can define tasks in plain English or use a visual builder to shape each step. The project offers two primary paths: a managed, hosted AutoGPT Platform that handles infrastructure and model access for a fee, and a self-hosting option that is free but requires users to provide their own infrastructure and model API keys. Agents can run on demand, on schedules, or from triggers, connecting to over 45 platforms and hundreds of AI models. It's presented as a tool to automate various functions, from executive operations and sales research to marketing campaign drafts and incident triage in engineering.
Firecrawl is an API and set of SDKs for turning any URL into Markdown, structured JSON, or a screenshot, handling JS rendering and proxy rotation itself. It's built for feeding AI agents and LLM pipelines clean web data, with dedicated search, crawl, map, and agent endpoints, an MCP server, and SDKs for nine languages.
anthropics/skills is Anthropic's public collection of Agent Skills for Claude — SKILL.md folders spanning creative, technical, enterprise, and document tasks, plus a spec and a template for building your own.
prompts.chat is an open-source library of prompts written for AI chat assistants, first released in December 2022 under the name Awesome ChatGPT Prompts. The GitHub project has since grown and now distributes prompts through a website, a CSV file, a Markdown file, and a Hugging Face dataset, alongside a self-hosting option, a CLI, an MCP server, and a Claude Code plugin.
prompts.chat is the largest open-source prompt library for AI, formerly called Awesome ChatGPT Prompts. It hosts curated prompts in CSV and Markdown, available as a public website, Hugging Face dataset, or self-hosted instance. The project supports multiple LLM providers including ChatGPT, Claude, Gemini, Llama, and Mistral. Self-hosting uses a Next.js setup wizard that configures authentication via GitHub, Google, or Azure AD, with PostgreSQL as the recommended database. CLI access, an MCP server, and a Claude Code plugin extend its reach into developer workflows. The codebase is MIT-licensed while prompt data falls under CC0. Its 166k GitHub stars make it an AI resource on the platform with 166k GitHub stars, and it has been cited by Harvard, Columbia, and Forbes.
Hugging Face Transformers (huggingface/transformers) is a Python library that centralizes model definitions for text, computer vision, audio, video, and multimodal machine learning, covering both inference and training. The README describes it as a pivot point compatible with training frameworks such as Axolotl, DeepSpeed, and PyTorch-Lightning, and inference engines such as vLLM, SGLang, and TGI, with more than 1M+ model checkpoints listed on the Hugging Face Hub.
JavaGuide là một repository tổng hợp kiến thức phỏng vấn Java và backend được duy trì bởi cộng đồng, bao phủ gần như toàn bộ các chủ đề mà một ứng viên backend cần nắm vững: từ Java core (collections, concurrency, JVM), hệ điều hành và mạng máy tính, cho đến cơ sở dữ liệu (MySQL, Redis), kiến trúc phân tán, và các công cụ devops như Docker, Git. Điểm mạnh rõ rệt là mỗi phần đều đi kèm bài tổng hợp câu hỏi phỏng vấn được đánh dấu "必看" (bắt buộc đọc), kèm theo phân tích source code chi tiết cho các cấu trúc dữ liệu phổ biến như HashMap, ConcurrentHashMap, và các topic nâng cao như AQS, JMM. Repository cũng mở rộng sang AI application development với một repo riêng (AIGuide) về LLM, Agent, RAG và MCP, phản ánh xu hướng hiện tại trong việc tuyển dụng backend. Nhược điểm là nội dung rất dàn trải — hơn 100 file markdown trải dài từ BigDecimal đến Elasticsearch, nên nếu bạn đã có nền tảng vững ở một số phần, việc lọc đúng nội dung cần ôn tập đòi hỏi thêm effort. Cá nhân tôi thấy đây là tài liệu tốt nhất để "làm quen" với toàn cảnh backend Java trước khi đi sâu vào từng chủ đề riêng lẻ, nhưng không nên dùng nó thay cho việc đọc source code thật hoặc tài liệu chính thức.
Langflow pairs a drag-and-drop canvas with editable Python components, letting you build an LLM-powered agent flow and ship it as an API, MCP server, or JSON export.
Dify is an open-source LLM app platform that packages a visual workflow/agent builder, RAG pipeline, model provider management, and app-level APIs into one self-hostable (or cloud) stack, aimed at teams who want to skip stitching those pieces together themselves.
The Agency is a collection of specialized AI agent personas — markdown prompt files with a defined identity, workflow, and success metrics — for Claude Code, Cursor, and other AI tools, installable through a CLI script or a native desktop app.
LangChain is a Python framework (MIT-licensed) for assembling LLM-powered apps and agents from standard components: model wrappers, prompts, retrieval, tools, and chain/graph orchestration handed off to LangGraph. It's for developers gluing together model providers and data sources, not for a single prompt-response call.
A GPL-3.0 GitHub repo compiling system prompts, tool definitions, and model names for 25+ AI products, from Cursor and Claude Code to lesser-known agents like Cluely and Qoder.
Shubhamsaboo/awesome-llm-apps is a substantial collection of over 100 open-source AI agents, agent skills, and Retrieval Augmented Generation (RAG) applications. Hand-built, end-to-end tested, and licensed under Apache-2.0, this Python-based repository supports a range of LLMs including Claude, Gemini, GPT, DeepSeek, Llama, and Qwen. It offers a wide array of examples, from single-file starter agents and advanced multi-agent teams to voice AI agents, generative UIs, autonomous game-playing agents, and various RAG implementations. The collection also includes LLM optimization tools, fine-tuning recipes, and crash courses on agent frameworks, providing practical, ready-to-deploy solutions for common LLM development challenges.
gstack is Garry Tan's open-source pack of 23 Claude Code skills and 8 power tools — a CEO, an eng manager, a QA lead, a security officer, and more — each a Markdown slash command that runs one stage of a sprint from idea to a shipped PR. It's MIT-licensed and free, documented as a roughly 30-second install for Claude Code, with setup support for OpenClaw, Cursor, and seven more agents.
llama.cpp is an open-source project at ggml-org/llama.cpp that implements LLM and VLM inference in plain C/C++ with no external dependencies, built on top of the ggml tensor library. The README documents support for x86 SIMD, Apple Silicon, RISC-V, and GPU backends including CUDA, HIP, MUSA, Vulkan, and SYCL, along with quantization down to 1.5-bit, a CLI, and an OpenAI-compatible server.
Microsoft's introductory course on building generative AI applications walks you through 21 structured lessons created by Microsoft Cloud Advocates. The curriculum alternates between conceptual 'Learn' lessons and hands-on 'Build' lessons, progressing from LLM fundamentals through advanced topics like RAG, AI agents, and fine-tuning. Each lesson includes a written explanation, video introduction, and working code samples in both Python and TypeScript. Basic Python or TypeScript knowledge is expected. The course supports Azure OpenAI, OpenAI API, Microsoft Foundry Models, and Foundry Local for fully offline execution. With over 50 language translations maintained via automated GitHub Actions, it's accessible to a global developer audience. MIT licensed with an active Discord community and developer forum for peer support.
MoneyPrinterTurbo is a self-hosted AI video generation pipeline that automates scriptwriting, stock footage selection, text-to-speech narration, subtitle rendering, and ffmpeg-based video assembly into HD MP4s. It ships with four interfaces (WebUI, API, CLI, AI Agent) and supports batch production. The tool is particularly attractive for creators who need volume at low per-video cost, since you supply your own API keys rather than paying per-export fees. Setup involves Python 3.11+, at least one LLM provider (many supported), and ffmpeg — Whisper captioning is optional but demands a multi-gigabyte model download. It's a legitimate workhorse for automated short video workflows, not a toy.
Browser Use provides a framework for AI agents to interact with web browsers, enabling automated tasks like form filling, data extraction, and QA testing. It offers both a Python library for programmatic control and a CLI for agent-driven workflows, supporting various LLMs and offering cloud services for scalability and advanced browser fingerprinting.
Cognee is an open-source AI memory platform that lets AI agents maintain persistent, context-aware memory across sessions. The platform combines vector embeddings with graph reasoning to turn raw documents into connected, queryable knowledge that agents can recall, update, and forget. It ships as a Python library with a CLI, Docker images, and official clients for Rust and TypeScript. Four operations handle the core workflow: remember, recall, forget, and improve. The platform can run entirely on Postgres using pgvector and a demo graph backend, or you can swap in dedicated graph databases like Neo4j, vector stores like LanceDB, and session caches like Redis. A Claude Code plugin brings persistent memory directly into that environment. Benchmarks on the BEAM long-context test show scores of 0.79 at 100K tokens, which drops to 0.67 at 10M tokens.
Impeccable is a JavaScript-based system designed to provide specific design guidance for AI coding agents, aiming to improve the quality of AI-generated frontend designs beyond common, repetitive patterns. Originating from Anthropic's `frontend-design` skill, it offers a single setup flow using `/impeccable init` to establish project-specific design context, including brand, audience, and anti-references. Developers can interact with their AI using 23 commands like `polish`, `audit`, and `critique`, forming a shared design vocabulary. The tool also incorporates 59 deterministic detector rules, which can run via a CLI or browser extension without an LLM or API key, alongside LLM-only critique checks to identify design anti-patterns. Impeccable integrates with a range of AI coding tools such as Cursor, Claude Code, GitHub Copilot, and Grok Build, offering various installation methods including a CLI installer, Git submodule, and direct plugin options. It also provides a standalone `npx impeccable detect` CLI for scanning codebases and URLs for design issues, and a design hook that surfaces findings during UI file edits in supported agents.
Unsloth brings local AI model management, training, and deployment into a single desktop and web interface. It targets developers and researchers who want to run and fine-tune large models on their own hardware without managing multiple toolchains. Three entry points exist: the Tauri desktop app (Unsloth Desktop), the browser-based Unsloth Studio, or the pip-installed Unsloth Core. The platform supports a wide model family including Llama, Gemma, Qwen, DeepSeek, and diffusion models, with claimed speedups of 2× and VRAM reductions of 70% during fine-tuning.
Firecrawl is an API and set of SDKs for turning any URL into Markdown, structured JSON, or a screenshot, handling JS rendering and proxy rotation itself. It's built for feeding AI agents and LLM pipelines clean web data, with dedicated search, crawl, map, and agent endpoints, an MCP server, and SDKs for nine languages.
herdr is a terminal agent multiplexer written in Rust that keeps every AI coding agent you're running visible in one place. Think tmux, but built around the fact that you're now juggling several coding agents at once — each pane shows the real terminal, with the agent's state read as blocked, working, or done.
MoneyPrinterTurbo is a self-hosted AI video generation pipeline that automates scriptwriting, stock footage selection, text-to-speech narration, subtitle rendering, and ffmpeg-based video assembly into HD MP4s. It ships with four interfaces (WebUI, API, CLI, AI Agent) and supports batch production. The tool is particularly attractive for creators who need volume at low per-video cost, since you supply your own API keys rather than paying per-export fees. Setup involves Python 3.11+, at least one LLM provider (many supported), and ffmpeg — Whisper captioning is optional but demands a multi-gigabyte model download. It's a legitimate workhorse for automated short video workflows, not a toy.
OpenWork is an open-source desktop application designed to centralize and share AI workflows, skills, and connected services across various AI agents. It acts as an alternative to proprietary solutions, allowing teams to collaborate on AI-driven tasks and manage access to models and plugins.
The `huggingface/speech-to-speech` project provides a low-latency, modular voice-agent pipeline, integrating Voice Activity Detection (VAD), Speech to Text (STT), Language Model (LLM), and Text to Speech (TTS) components. It exposes this functionality through an OpenAI Realtime-compatible WebSocket API, allowing for a fully local or hybrid stack where every component is swappable. This Python-based system supports various backends for each stage, including local inference via Transformers or `mlx-lm`, and self-hosted or provider APIs for LLMs. Thousands of Reachy Mini robots rely on this system as their conversation backend in production, according to the README.
The largest collection of PyTorch image encoders / backbones. Including train, eval, inference, export scripts, and pretrained weights -- ResNet, ResNeXT, EfficientNet, NFNet, Vision Transformer (ViT), MobileNetV4, MobileNet-V3 & V2, RegNet, DPN, CSPNet, Swin Transformer, MaxViT, CoAtNet, ConvNeXt, and more
中文分词 词性标注 命名实体识别 依存句法分析 成分句法分析 语义依存分析 语义角色标注 指代消解 风格转换 语义相似度 新词发现 关键词短语提取 自动摘要 文本分类聚类 拼音简繁转换 自然语言处理
OpenPose: Real-time multi-person keypoint detection library for body, face, hands, and foot estimation
💫 Industrial-strength Natural Language Processing (NLP) in Python
Visualizer for neural network, deep learning and machine learning models
An LLM-powered knowledge curation system that researches a topic and generates a full-length report with citations.
Data science Python notebooks: Deep learning (TensorFlow, Theano, Caffe, Keras), scikit-learn, Kaggle, big data (Spark, Hadoop MapReduce, HDFS), matplotlib, pandas, NumPy, SciPy, Python essentials, AWS, and various command lines.
Scalable, Portable and Distributed Gradient Boosting (GBDT, GBRT or GBM) Library, for Python, R, Java, Scala, C and more. Runs on single machine, Hadoop, Spark, Dask, Flink and DataFlow
Integrate cutting-edge LLM technology quickly and easily into your apps
Implementation of Vision Transformer, a simple way to achieve SOTA in vision classification with only a single transformer encoder, in Pytorch
[NeurIPS'23 Oral] Visual Instruction Tuning (LLaVA) built towards GPT-4V level capabilities and beyond.
