The most valuable
repos on GitHub
Curated AI frameworks, Claude skill repos, MCP servers, and open source projects you can deploy in minutes — all free.
Most valuable AI repositories
Frameworks, tools, and libraries every AI builder should know — sorted by category
Awesome Claude Prompts
Official Anthropic cookbook — real code examples for every Claude API feature. RAG, tool use, streaming, vision, PDF analysis. Each notebook is copy-paste ready.
Prompt Engineering Guide
The most comprehensive guide to prompt engineering. Covers chain-of-thought, few-shot, RAG, ReAct, and every major prompting technique with examples across all major models.
LangChain
The most popular framework for building LLM-powered applications. Chains, agents, memory, RAG — LangChain has abstractions for everything. Connects to 100+ LLMs and 50+ vector stores.
LlamaIndex
The best framework specifically for RAG and data-connected LLM apps. Superior to LangChain for document Q&A. Handles chunking, indexing, querying, and synthesis intelligently.
CrewAI
Build multi-agent systems where AI agents collaborate like a human team. Define roles, assign tasks, set goals — CrewAI orchestrates the collaboration automatically.
AutoGen (Microsoft)
Microsoft's framework for multi-agent conversation systems. Agents can code, execute, and iterate without human intervention. Best for automated research and coding tasks.
DSPy
Stanford's framework for algorithmically optimising LLM prompts and pipelines. Instead of writing prompts manually, DSPy compiles them from examples. The future of prompt engineering.
Semantic Kernel
Microsoft's enterprise AI SDK for .NET, Python, and Java. Integrates LLMs into enterprise applications. Best choice if your team uses C#/.NET or builds enterprise software.
Haystack
Enterprise-grade NLP framework for building search and RAG pipelines. Better suited for production search systems than LangChain. Used at scale by large companies.
Phidata
Build full-stack AI agents with memory, knowledge, and tools. Ships with pre-built financial agents, research agents, and coding assistants. Production-ready from day one.
Weaviate
Open-source AI-native vector database with hybrid search (vector + keyword). Built-in ML model integration, GraphQL API, and real-time updates. Enterprise-scale tested.
Milvus
The most scalable open-source vector database. Handles billions of vectors. Used by Salesforce, eBay, and hundreds of enterprises. When ChromaDB is too small for your scale.
RAGAS
The standard framework for evaluating RAG pipelines. Measures faithfulness, answer relevancy, context recall, and precision automatically. Essential before shipping any RAG product.
DeepEval
pytest for LLM applications. Write unit tests for your AI outputs — check for hallucination, bias, toxicity, and correctness. CI/CD integration for LLM quality assurance.
LiteLLM
Universal LLM proxy — one API format for 100+ models. Switch between OpenAI, Claude, Gemini, Llama, and more with zero code changes. Built-in cost tracking and rate limiting.
Instructor
Get structured, validated data from LLMs using Pydantic models. Ask an LLM a question and get back a proper Python object, not a raw string. Eliminates JSON parsing headaches.
Outlines
Guaranteed structured generation from local models. Force any LLM to output valid JSON, follow a regex, or match an enum. No retries, no validation loops — mathematically guaranteed.
Guidance (Microsoft)
Control LLM generation with structured templates. Mix regular text, loops, conditionals, and LLM calls in one template. Far more predictable than free-form prompting.
Marvin
AI functions and classifiers as Python decorators. Add @ai_fn to any function and it works with LLMs. Classify, extract, generate, and transform data with one line of code.
ComfyUI
The most powerful node-based UI for Stable Diffusion, Flux, and video models. Visual workflow builder for any image/video generation pipeline. Supports every model including Wan 2.1 and LTX.
InvokeAI
Professional local image generation studio. Better UI than Automatic1111, supports Flux and SDXL, has inpainting/outpainting, canvas tools, and a workflow builder. The best all-in-one local image tool.
Fish Speech
State-of-the-art open-source TTS model that clones any voice from 10 seconds of audio. Supports 8 languages including Chinese and Hindi. Runs locally, produces ElevenLabs-quality output.
OpenVoice
Zero-shot voice cloning from a single audio sample. Just 5 seconds of any voice and you can generate unlimited speech in that voice. Used for dubbing, content creation, and accessibility.
MetaVoice
Open-source 1.2B parameter TTS model. Supports emotion control and voice cloning. Apache 2.0 license means commercial use is free. Best for emotionally expressive speech.
Khoj
Your personal AI that searches across all your notes, documents, and the web. Indexes Obsidian, Notion, Org-mode files. Ask questions across your entire knowledge base.
PrivateGPT
Ask questions to your documents — 100% private, no internet required. Upload PDFs, Word docs, text files and chat with them locally using Ollama or LlamaCpp. Zero data leaves your machine.
AnythingLLM
All-in-one private ChatGPT alternative. Import documents, websites, YouTube transcripts and chat with them. Supports 30+ LLM providers and multiple vector databases. Desktop app available.