The best open-source AI projects, all in one place
A curated directory of must-know open-source AI projects on GitHub — LLM engines, agent frameworks, RAG tools, image and speech models — with stars, forks, and language info.
Browse by category
Chat & Assistants
5Open-source chat interfaces and assistant apps you can self-host and customize.
Inference & Serving
9Engines and gateways for running LLMs locally or serving them at scale.
Agents & Workflows
9Frameworks and platforms for building autonomous agents and agentic workflows.
RAG & Knowledge
7Retrieval-augmented generation engines, vector databases, and document intelligence.
Training & Fine-tuning
7Tools for fine-tuning, reinforcement learning, and distributed training of LLMs.
Image & Video Generation
6Diffusion UIs, pipelines, and models for generating images and video.
Speech & Audio
7Speech recognition, voice cloning, and text-to-speech models and toolkits.
Libraries & Tools
5Core ML libraries, model hubs, tokenizers, and observability tooling.
Open-source chat interfaces and assistant apps you can self-host and customize.
User-friendly self-hosted AI interface supporting Ollama, OpenAI-compatible APIs, and more.
Zero-config AI chat assistant with 100+ models, available on Web, iOS, macOS, Android, and Linux.
Open-source AI chat framework with agents, plugin ecosystem, and multi-model support.
Enhanced multi-provider ChatGPT clone with agents, MCP, artifacts, and model switching.
The open-source codebase powering HuggingChat, deployable with any inference backend.
Engines and gateways for running LLMs locally or serving them at scale.
Run large models locally with a single command — LLMs, vision, and embeddings on any machine.
LLM inference in C/C++ with quantization — runs models on CPU, GPU, and edge devices.
High-throughput, memory-efficient inference and serving engine for LLMs with PagedAttention.
AI gateway to call 100+ LLM APIs in OpenAI format with cost tracking, guardrails, and load balancing.
Run any model — LLMs, vision, voice, image, video — on any hardware, no GPU required.
Desktop app for local LLMs with text, vision, tool-calling, and OpenAI-compatible APIs.
High-performance serving framework for large language and multimodal models.
Distribute and run LLMs with a single cross-platform executable file.
Unified, production-ready inference API for open-source LLMs, speech, and multimodal models.
Frameworks and platforms for building autonomous agents and agentic workflows.
Accessible autonomous AI agents: build and run agents that pursue goals across multi-step workflows.
Build agentic workflows and RAG pipelines in one collaborative workspace, cloud or self-hosted.
Framework for building LLM-powered agents and applications with composable components.
Open-source Python library for AI browser agents with stealth browsers and a hosted agent API.
AI-driven development platform: autonomous agents that write code, run commands, and open PRs.
Microsoft framework for agentic AI: orchestrate multi-agent conversations and workflows.
Framework for orchestrating role-playing autonomous AI agents that collaborate on complex tasks.
Self-driving agent platform for building, running, and monitoring multi-agent systems.
OpenAI's educational framework for ergonomic, lightweight multi-agent orchestration.
Retrieval-augmented generation engines, vector databases, and document intelligence.
Agent-native RAG engine with deep document understanding, from retrieval to context infrastructure.
All-in-one local-first AI agent and document chat app with multi-user support.
Document processing platform for AI: ingest, index, and query your data with LLMs.
Turns continuous content streams into search and monitoring with durable ingestion and hybrid search.
High-performance, massive-scale vector database and search engine written in Rust.
Open-source embedding database built for AI-native search and retrieval.
Open-source RAG tool for chatting with your documents, with clean citations.
Tools for fine-tuning, reinforcement learning, and distributed training of LLMs.
Fine-tune and run LLMs and diffusion models 2x faster with far less memory.
Unified efficient fine-tuning of 100+ LLMs and VLMs with a no-code web UI.
Deep learning optimization library that makes distributed training and inference easy and efficient.
State-of-the-art parameter-efficient fine-tuning: LoRA, Prefix Tuning, and more.
Train transformer language models with reinforcement learning: SFT, DPO, PPO, and GRPO.
NVIDIA's framework for training large transformer models at scale.
Streamline fine-tuning of various AI models with YAML-based configs.
Diffusion UIs, pipelines, and models for generating images and video.
The most widely used browser interface for Stable Diffusion image generation.
Powerful modular diffusion GUI and backend with a graph/nodes interface.
Image generation focused on prompting — Simplified, offline, and open source.
Creative engine for Stable Diffusion with a professional studio workflow.
Stability AI's official generative model repositories including SDXL and SV3D.
Systematic open-source framework for large-scale video generation models.
Speech recognition, voice cloning, and text-to-speech models and toolkits.
Robust speech recognition model trained on large-scale weak supervision.
Few-shot voice cloning: train a good TTS model from just one minute of voice data.
Deep learning toolkit for text-to-speech, battle-tested in research and production.
Generative speech model optimized for daily dialogue, in Chinese and English.
Text-prompted generative audio model: speech, music, and sound effects.
State-of-the-art open-source text-to-speech with multilingual voice cloning.
Whisper transcription with CTranslate2 — up to 4x faster with less memory.
Core ML libraries, model hubs, tokenizers, and observability tooling.
The model-definition framework for state-of-the-art ML models across text, vision, and audio.
Build and share machine learning demos and apps entirely in Python.
Open-source LLM engineering platform: tracing, evals, and observability for agents.
State-of-the-art diffusion models for image, video, and audio generation in PyTorch.
Fast BPE tokenizer for use with OpenAI's models.