Awesome Self-Hosted AI

Open-source AI you can run on your own hardware or server.

303 projects across local LLMs, chat interfaces, agents, MCP servers, RAG, speech and image generation. A weekly job checks every project against the GitHub API, and anything archived, deleted or silent for over a year moves to UNMAINTAINED.md.

303 projects | 302 active in the last 90 days | 10.5M combined stars | 0 unmaintained | refreshed 2026-10-31

See the live health table for stars, last push and license per project.

Contents

• Agents and Automation (27)
• Chat Interfaces (11)
• Coding Assistants (22)
• Document AI and OCR (10)
• Fine-Tuning and Training (10)
• Home and Personal AI (13)
• Image and Video Generation (12)
• LLM Runtimes and Inference (37)
• MCP Servers and Gateways (19)
• Model Gateways and Routers (11)
• Observability and Evals (20)
• RAG and Vector Databases (24)
• Search and Knowledge (13)
• Speech and Audio (15)
• Vision and Multimodal (5)
• Voice Agents (5)
• Rising (25)
• Source-Available (24)

Agents and Automation

• Agent Development Kit - Python toolkit for building and evaluating LLM agents, with multi-model support.
• AgentScope - Multi-agent framework with tracing, memory and tool use for building and deploying agents.
• Agno - Python framework for building multi-agent systems with memory, knowledge, and a runtime for agent apps.
• Botpress - Platform for building and self-hosting conversational AI chatbots and agents with a visual flow builder.
• browser-use - Library that lets LLM agents drive a real browser to complete web tasks.
• CrewAI - Framework for orchestrating role-based teams of AI agents inside your own Python environment.
• Cua - Sandboxes, SDKs and benchmarks for running computer-use agents against desktop environments.
• DeerFlow - Agent harness that coordinates subagents, sandboxed execution, and memory for long-running tasks.
• Forge - Python framework for tool calling and multi-step workflows on self-hosted LLMs.
• Hermes Agent - Autonomous agent that runs server-side, keeps persistent memory and refines its behavior over time.
• Julep - Server platform for stateful agent workflows with memory, tools, branching and long-running tasks.
• Langflow - Visual builder for AI agents and RAG workflows in Python, with each flow exposed as an API.
• LangGraph - Library for building stateful, long-running agent workflows as graphs with checkpointing.
• Letta - Server for stateful agents that keep long-term memory and run as persistent services.
• mem0 - Memory layer for AI assistants that stores and retrieves user facts across sessions.
• Microsandbox - Local microVM runtime that isolates AI agent and untrusted workloads in sandboxed execution.
• Open Interpreter - Natural-language interface that lets an LLM run code and shell commands on your own computer.
• open-multi-agent - Self-hosted TypeScript agent runtime with durable approvals and verifiable run records.
• OpenManus - General-purpose agent framework that runs tools and a browser on your own machine.
• OpenShell - Runtime for autonomous agents that enforces policy-driven execution boundaries, with a CLI.
• Paperclip - Open-source MIT app for coordinating AI agents across projects, tasks and workflows.
• Pydantic AI - Type-safe Python agent framework with model-agnostic tool calling and structured outputs.
• Sim - Visual workflow builder for agents and LLM pipelines that can be self-hosted with Docker.
• Steel Browser - Open-source browser sandbox with an API that lets AI agents run web automation sessions.
• Trigger.dev - TypeScript background job framework for long-running AI tasks with retries and run monitoring.
• UI-TARS Desktop - Desktop app that runs ByteDance's multimodal GUI agent to control a local computer.
• Zep - Memory service that stores and retrieves facts and conversation history for AI agents over time.

Chat Interfaces

• AnythingLLM - Self-hosted app for chatting with documents using local or cloud models and built-in RAG.
• big-AGI - Multi-model chat app with personas and tools that can be self-hosted for power users.
• Chatbox - Cross-platform desktop and web client for chatting with local and remote LLM APIs.
• Cherry Studio - Desktop client for chatting with many LLM providers, with knowledge bases and local document search.
• HuggingChat (self-hosted) - Self-hostable chat web app that connects to OpenAI-compatible model endpoints.
• Jan - Open-source desktop app for running local models and chatting with cloud providers.
• LibreChat - Multi-provider chat UI with plugins, file uploads, and agents that runs in Docker on your server.
• NextChat - Lightweight web chat client for many LLM providers, deployable as a single container.
• Perplexica - AI answering search engine that cites web sources and runs with local or cloud models.
• SillyTavern - Customizable web frontend for chatting and roleplaying with locally hosted or remote LLMs.
• text-generation-webui - Gradio web UI for running local models through llama.cpp, Transformers, and ExLlama backends.

Coding Assistants

• Archon - YAML-defined workflow engine that makes coding agent processes repeatable from planning to validation.
• Claude Code Local - MLX server exposing the Anthropic Messages API so the Claude Code CLI can run against local models on Apple Silicon.
• Cline - VS Code agent that plans and edits code using model API keys you supply.
• Codex CLI - Local terminal coding agent that reads, edits, and runs code inside your repository.
• Continue - IDE extension for autocomplete, chat, and agentic edits against local or hosted models.
• DeepSeek-Reasonix - Terminal coding agent built around DeepSeek models, tuned for prefix-cache stability.
• Gemini CLI - Open-source terminal agent for coding and shell tasks, built on Google Gemini models.
• goose - On-machine coding agent that runs shell, file, and extension tools with any supported model.
• Goose - On-machine development agent that edits files and runs tasks, with extensible tool support.
• gptme - Terminal agent with local tools that writes code, runs shell commands and browses the web.
• Kilo Code - Open-source coding agent for VS Code and the terminal, with bring-your-own model keys.
• Mistral Vibe - Lightweight command-line coding agent from Mistral for local development workflows.
• Open Code Review - Code review CLI that combines deterministic analysis passes with LLM agents.
• opencode - Terminal coding agent that works with whichever model provider you configure.
• OpenCode - Terminal-native autonomous coding agent for editing and running code from the command line.
• OpenHands - Platform for AI software agents that edit code, run commands, and browse inside a sandbox.
• Pi - Modular terminal coding agent toolkit with a unified model layer and TUI and web front ends.
• PR-Agent (Qodo) - AI pull request reviewer that runs on GitHub, GitLab, Bitbucket, and Azure DevOps.
• Qwen Code - Terminal coding agent tuned for Qwen models and compatible with other OpenAI-style APIs.
• screenshot-to-code - Turns screenshots, mockups, and Figma designs into frontend code using vision-language models.
• SWE-agent - Software engineering agent that resolves GitHub issues by editing code and running commands in a container.
• T3 Code - Minimal web and desktop GUI for driving coding agents such as Codex and Claude Code.

Document AI and OCR

• Docling - Parses PDF, DOCX, PPTX and HTML into structured Markdown or JSON, with OCR and table recovery.
• GLM-OCR - Open OCR model and code from Z.ai for text, table and formula recognition in document images.
• kordoc - Converts HWP, HWPX, PDF and Office files to Markdown, with a CLI and an MCP server.
• LiteParse - Fast, local, model-free document parser for extracting text and layout from documents.
• Marker - Converts PDFs and other documents to Markdown, with table extraction and optional LLM cleanup.
• MarkItDown (Microsoft) - Command-line and Python tool that converts Office files, PDFs, and images into Markdown.
• OpenContracts - Self-hosted platform for annotating legal documents with semantic search and contract analysis.
• OpenMed - Local-first healthcare app for de-identifying clinical text and extracting medical entities.
• PaddleOCR - OCR toolkit with text detection, recognition, and layout analysis that can run as a local service.
• Unstructured - ETL toolkit and API that parses complex documents into structured elements for language model pipelines.

Fine-Tuning and Training

• Axolotl - Config-driven toolkit for full and LoRA fine-tuning and preference training on your own GPUs.
• H2O LLM Studio - No-code GUI for fine-tuning LLMs with supervised training and reward modeling.
• LLaMA-Factory - Unified framework for fine-tuning 100+ LLMs and VLMs with LoRA, QLoRA and RLHF methods.
• LLaMA-Factory - Web UI and CLI for fine-tuning LLMs with SFT, DPO, ORPO and KTO.
• MergeKit - Command-line tools for merging pretrained language models using configurable merge recipes.
• nnU-Net - Command-line segmentation pipeline that configures and trains its own model for a medical imaging dataset.
• OneTrainer - GUI application for training diffusion image models such as SD 1.5, SDXL and FLUX.
• Oumi - Open platform with a CLI for fine-tuning, evaluating and deploying open LLMs and vision-language models.
• Soup - Fine-tunes LLMs from one YAML file, with layer streaming for GPUs with limited memory.
• TRL - Library for post-training language models with SFT, DPO and GRPO on top of Transformers.

Home and Personal AI

• AI Novel Writer - Local-first desktop workbench for drafting long fiction with characters, worldbuilding and chapter plans.
• AionsHome - Self-hosted AI companion with long-term memory, voice interaction and camera-based vision.
• AIRI - Self-hosted AI companion platform with Live2D characters and real-time voice conversation.
• AstrBot - Agent framework and bot server that connects LLMs to many messaging platforms.
• DeepTutor - Multi-agent tutoring system that adapts lessons to a learner through interactive workflows.
• Home Assistant - Home automation platform with local voice assistants and LLM conversation agents.
• Khoj - Self-hostable personal assistant for chat, search, and automations over your notes and documents.
• llmfit - Terminal tool that scores which local models fit your CPU, GPU and RAM before you download them.
• nanobot - Lightweight personal AI agent with a built-in web UI, model routing, MCP support, and chat channels.
• Octop - Self-hosted AI assistant for several users, running multiple agents side by side.
• OGAM - Offline AI app for phone and Mac that chats, sees, speaks and generates images with local models.
• OpenClaw - Local-first personal AI assistant that links chat channels to agents running on your own machine.
• RuView - Software and firmware that senses presence, pose and vitals from commodity WiFi signals without video.

Image and Video Generation

• AI Toolkit - Scripts and UI for training LoRAs and fine-tunes on diffusion and video models.
• ComfyUI - Node-based interface and engine for running diffusion image and video pipelines locally.
• ComfyUI - Node-based editor and server for building and running diffusion image and video workflows locally.
• DiffSynth-Studio - Diffusion engine for image and video generation with memory-optimized local inference.
• Diffusers - Library of diffusion pipelines and schedulers for image, video, and audio generation in PyTorch.
• InvokeAI - Creative image generation studio with a browser UI, canvas editing, and model management.
• iris.c - Pure C inference engine for the FLUX.2 image generation model.
• Mochi - Open text-to-video model released with inference code for running on local GPUs.
• MoneyPrinterTurbo - Pipeline that produces short videos from a topic, covering script, footage, and voiceover.
• MOSS-Image - Official code for MOSS-Image, a 5B-parameter diffusion transformer for high-quality image generation.
• sd-scripts - Training and fine-tuning scripts for Stable Diffusion and related image models.
• SD.Next - WebUI for image and video generation with many diffusion models and multiple hardware backends.

LLM Runtimes and Inference

• Aphrodite Engine - Large-scale LLM inference engine with a REST API and LoRA support, forked from vLLM.
• Colibri - Zero-dependency C engine that runs large mixture-of-experts models locally by streaming experts from disk.
• distributed-llama - Runs one LLM across several home devices over the network, so adding machines speeds up inference.
• ds4 - Native local inference engine for DeepSeek 4 Flash and Pro models on Metal, CUDA and ROCm GPUs.
• ExecuTorch - PyTorch runtime and toolchain for deploying models to mobile, embedded and edge hardware.
• ExLlamaV3 - Inference library for running quantized LLMs efficiently on consumer NVIDIA GPUs.
• exo - Peer-to-peer system that pools devices on a local network to run one larger model.
• gemma.cpp - Lightweight standalone C++ engine for running Google Gemma models locally.
• GPUStack - GPU cluster manager that places and runs model servers such as vLLM and SGLang across your machines.
• ik_llama.cpp - Llama.cpp fork with extra state-of-the-art quantization types and faster hybrid GPU and CPU inference.
• KTransformers - Inference framework that offloads MoE experts to the CPU so large models fit on modest GPUs.
• Lemonade - Local LLM server with an OpenAI-compatible API that uses GPU and NPU acceleration.
• Llama Stack - Server with unified APIs for inference, RAG, agents, and safety that runs Llama and other models.
• llama-swap - Lightweight proxy that swaps between llama.cpp, vLLM, and other model servers per request.
• llama.cpp - C and C++ inference engine for running GGUF models on CPUs and GPUs across many backends.
• llamafile - Single-file executables that bundle a model with llama.cpp so it runs locally on most computers.
• llm-d - Kubernetes-native framework for distributed LLM inference that spreads serving across cluster nodes.
• LMDeploy - Toolkit for quantizing, deploying and serving LLMs, with 4-bit inference and an API server.
• LocalAI - OpenAI-compatible server that runs text, image, and audio models on local hardware.
• Magnitude - Inference engine that tunes kernels on your own hardware to find the fastest way to run each model.
• mistral.rs - Rust LLM inference engine with an OpenAI-compatible server and quantization support.
• MLC LLM - Compiler-based LLM deployment that runs models natively on phones, browsers, and desktop GPUs.
• mlx-lm - Run and fine-tune language models on Apple silicon with the MLX framework.
• MNN - Lightweight inference runtime from Alibaba for running and training neural networks on devices.
• NVIDIA Dynamo - NVIDIA datacenter-scale distributed inference server that orchestrates engines such as vLLM and SGLang.
• Ollama - Local model runner with a CLI and HTTP API for pulling and serving open-weight LLMs.
• omlx - Local LLM server for Apple Silicon that batches requests, caches to SSD, and runs from the macOS menu bar.
• RamaLama - Developer CLI that serves local models inside containers, detecting the GPU and pulling a matching image.
• SGLang - Serving framework for LLMs and multimodal models with structured generation and prefix caching.
• shimmy - Single-binary Rust inference server with an OpenAI-compatible API for GGUF and SafeTensors models.
• TensorRT-LLM - An open inference backend and runtime for serving LLMs efficiently on NVIDIA GPUs.
• Text Embeddings Inference (Hugging Face) - Inference server for embedding, reranking, and classification models with token-based batching.
• Triton Inference Server - Model server that runs models from many frameworks on GPUs and CPUs with dynamic batching.
• Unsloth - Training toolkit that speeds up LoRA and full fine-tuning of open LLMs on your own GPUs.
• vLLM - High-throughput LLM serving engine with paged attention and an OpenAI-compatible server.
• vLLM Production Stack - Kubernetes deployment stack for vLLM that adds automated scaling and monitoring around the engine.
• Xinference - Model serving platform that exposes LLM, embedding, and speech models through one API.

MCP Servers and Gateways

• agentgateway - Rust proxy for agent traffic that supports MCP and A2A protocols.
• AI Memory - Rust memory server and MCP client that gives coding CLIs long-term recall with full-text search.
• AnythingMCP - Turns REST, GraphQL, SOAP and SQL APIs into MCP tools behind one gateway.
• Chrome DevTools MCP - MCP server that lets coding agents control and inspect a local Chrome instance via DevTools.
• cocoindex-code - Local AST-aware code search CLI and MCP server backed by an embedded SQLite index.
• codebase-memory-mcp - C-based MCP server that indexes repositories into local type-resolved knowledge graphs.
• DesktopCommander MCP - MCP server that gives AI clients terminal control, file search, and diff-based file editing.
• Docker MCP Gateway - Runs MCP servers in isolated containers and exposes them to clients through one gateway.
• FastMCP - Pythonic framework for building MCP servers and clients.
• GitHub MCP Server - Official MCP server exposing GitHub repositories, issues, and pull requests to AI clients.
• Korean Law MCP - MCP server and CLI for querying Korea's official statutes, precedents and administrative rules.
• MCP for Blender - Blender add-on and MCP server that lets LLM agents create and modify 3D scenes.
• mcp-context-forge - Gateway and registry that federates MCP and REST services with auth, rate limits, and an admin UI.
• mcp-proxy (sparfenyuk) - Bridges stdio MCP servers to SSE or streamable HTTP transports and back.
• mcp-proxy (TBXark) - Aggregates several MCP servers behind one endpoint with config-driven routing.
• mcp-use - Framework for building MCP clients and agents, with Python and TypeScript SDKs.
• Mobile MCP - MCP server that automates and inspects iOS and Android simulators, emulators, and real devices.
• Playwright MCP - MCP server giving LLM clients browser control through Playwright accessibility snapshots.
• Unla - Gateway that turns existing HTTP APIs and MCP servers into MCP endpoints with a web console.

Model Gateways and Routers

• Arch Gateway - Edge proxy built on Envoy that routes and guards LLM traffic across model providers.
• Bifrost - Go-based LLM gateway with failover, load balancing and an OpenAI-compatible API you host yourself.
• Claude Code Router - Local router that sends coding-agent requests to different model providers based on rules.
• CliRelay - Self-hosted gateway that gives coding CLIs one API across several model providers.
• Envoy AI Gateway - Envoy-based gateway that routes and secures traffic to multiple LLM providers.
• Headroom - Proxy that compresses tool outputs, logs and retrieved chunks to cut token use for AI agents.
• Higress (Alibaba) - AI-native API gateway that adds model routing and AI plugins on top of Envoy.
• LLM (Simon Willison) - Command-line tool for prompting many local and hosted models through one interface.
• New API - AI model gateway that fronts many upstream providers behind one key with quotas and billing.
• optillm - Optimizing inference proxy that adds routing, load balancing and failover across multiple LLM providers.
• Switchyard - Rust proxy that routes traffic across LLM providers while keeping OpenAI and Anthropic API compatibility.

Observability and Evals

• agentsview - Local-first search, analytics, and token and cost statistics for coding agent sessions.
• Aim - Self-hosted experiment tracker with a web UI for comparing very large numbers of training runs.
• Argilla - Data annotation and feedback platform for curating datasets and reviewing LLM outputs by hand.
• ChainForge - Visual node-based environment for comparing prompts and model outputs across many LLMs.
• DeepEval - Pytest-style evaluation framework for LLM apps with metrics for RAG and agent outputs.
• Future AGI - Self-hostable platform for tracing, evals, simulations and guardrails across LLM and agent apps.
• Garak (NVIDIA) - Command-line LLM vulnerability scanner that probes models for prompt injection, leakage and hallucination.
• Harbor - Framework for running agent evaluations and building reinforcement-learning environments from the CLI.
• Helicone - LLM observability gateway that logs requests, costs, and latency through a proxy you host.
• iFixAi - Command-line agent auditing framework that runs adversarial checks and produces scored reports.
• Inspect AI - Framework for evaluating LLMs and agents with solvers, scorers, and sandboxed tool execution.
• LangWatch - LLM monitoring and evaluation platform with tracing, datasets, and prompt management for AI apps.
• LM Evaluation Harness - Framework for running standard benchmark suites against language models locally.
• MLflow - ML lifecycle platform with tracing, prompt versioning, and evaluation for LLM applications.
• OpenLIT - OpenTelemetry-native observability for LLM apps, with tracing, cost tracking and GPU metrics.
• OpenLLMetry - OpenTelemetry instrumentation for LLM and vector database calls, exporting to any backend.
• Opik - Open-source LLM tracing and evaluation platform with a self-hostable Docker deployment.
• Pezzo - Open-source LLMOps platform for prompt versioning, delivery and observability.
• Promptfoo - CLI and library for evaluating prompts, models, and RAG outputs with test assertions.
• SkillSpector (NVIDIA) - Security scanner that checks agent skills and MCP tools for prompt injection and exfiltration risks.

RAG and Vector Databases

• Chroma - Embedding database for LLM apps with a single-node server and Python and JS clients.
• Cognee - Knowledge engine that builds graph and vector memory from documents for agents.
• Graphiti - Framework for building temporal knowledge graphs from agent and user data for retrieval.
• Haystack - Python framework for building search and RAG pipelines from swappable components.
• HelixDB - Graph database with built-in vector and full-text search, written in Rust on object storage.
• Hindsight - Self-hosted long-term memory service for AI agents with integrations for common agent frameworks.
• Infinity (AI Database) - AI-native database server offering hybrid dense, sparse, and full-text search for LLM applications.
• LanceDB - Embedded multimodal vector database that stores tables on local disk or object storage.
• LightRAG - Graph-based retrieval system that indexes entities and relations for faster RAG queries.
• LlamaIndex - Data framework that connects private documents to LLMs through indexes and retrievers.
• MaxKB - Knowledge base and agent platform for RAG chatbots and workflow automation over your own documents.
• MemPalace - Local AI memory system that stores and retrieves context for long-running agent workflows over MCP.
• Milvus - Distributed vector database built for large-scale embedding search with hybrid filtering.
• MindsDB - Federated SQL layer that connects databases and apps to AI models for querying and in-database ML.
• OpenViking - Context database that unifies agent memory, knowledge retrieval and skills as a virtual filesystem.
• pgvector - Vector similarity search extension for PostgreSQL with exact and approximate nearest-neighbor indexes.
• pgvectorscale - PostgreSQL extension adding DiskANN-based vector indexes for faster approximate nearest neighbor search.
• PrivateGPT - Offline document chat API with ingestion, embeddings, and an OpenAI-style interface.
• Qdrant - Vector similarity search engine in Rust with payload filtering and REST and gRPC APIs.
• Quivr - Open-source RAG app for chatting with your documents, runnable on your own infrastructure.
• RAGFlow - Document-understanding RAG engine with chunking templates, citations, and agent workflows.
• Supermemory - Memory engine and app for agents that can run fully locally, with context-aware storage and retrieval.
• Vespa - Search and serving engine combining vector and keyword retrieval with real-time indexing.
• zvec - In-process vector database library for fast similarity search without running a server.

Search and Knowledge

• Agent-Reach - CLI and search layer that lets agents pull content from Reddit, X, YouTube, GitHub and more.
• Codegraph - Local pre-indexed code knowledge graph that cuts token use and tool calls for coding agents.
• Crawl4AI - Open-source web crawler that turns pages into clean Markdown for retrieval and agent pipelines.
• DB-GPT - Self-hosted assistant for private knowledge and databases, with data-aware chat and RAG workflows.
• DocsGPT - Self-hosted platform for building document-grounded assistants and agents with enterprise search.
• Firecrawl - Self-hostable web data API that scrapes and crawls sites into clean markdown or JSON for LLMs.
• gpt-researcher - Autonomous research agent that searches the web and writes cited reports on any topic.
• LLM Wiki - Desktop app that turns your documents into a linked wiki maintained by an LLM.
• Open Notebook - Self-hosted alternative to NotebookLM for organizing sources, notes and chats over your documents.
• OpenResearch - Local-first workspace that turns coding agents into research agents for literature review.
• SearXNG - Privacy-respecting metasearch engine you run yourself, aggregating results from many search services.
• SurfSense - Self-hosted workspace for organizing and querying team knowledge, in the style of a notebook app.
• WeKnora - Knowledge platform that turns documents into a RAG index, a reasoning agent, and a maintained wiki.

Speech and Audio

• faster-whisper - Whisper reimplementation on CTranslate2 that transcribes faster with lower memory use.
• GPT-SoVITS - Few-shot voice cloning and text-to-speech system with a web interface, trained on short samples.
• Kokoro-FastAPI - OpenAI-compatible FastAPI server wrapping the Kokoro text-to-speech model.
• Meetily - Local meeting assistant that transcribes audio in real time and writes summaries on your machine.
• Moshi - Full-duplex speech-to-speech model and dialogue system for real-time spoken conversation.
• Piper - Fast local neural text-to-speech engine that runs well on small single-board computers.
• sherpa-onnx - Speech toolkit on ONNX Runtime for offline ASR, TTS and speaker tasks across many platforms.
• Silero VAD - Voice activity detector model that runs on CPU from a small ONNX file.
• Speaches - OpenAI-compatible server for transcription and speech synthesis on your own hardware.
• TypeWhisper - Local speech-to-text app for macOS that runs models on-device, with optional cloud.
• Voicebox - Local voice studio for cloning voices, generating multilingual speech, and dictating text.
• VoiceStudio - Local voice platform covering multi-engine text-to-speech, voice cloning, and video dubbing.
• Whisper - Reference speech recognition model and inference code for multilingual transcription.
• whisper.cpp - C and C++ port of Whisper for fast transcription on CPUs, Apple Silicon, and GPUs.
• WhisperX - Whisper pipeline that adds word-level timestamps and speaker diarization.

Vision and Multimodal

• CVAT - Web-based annotation tool for labeling images and video used to train computer vision models.
• JoyAI-Image - Multimodal foundation model for image understanding and text-to-image generation.
• MiniCPM-V - Multimodal model family for image, video and document understanding, with small variants for local use.
• modlens - Vision bridge that gives text-only LLM tools image understanding through a plugin.
• WeVisDoc - End-to-end document parser for page images, fine-tuned from Qwen3-VL-2B-Instruct.

Voice Agents

• LiveKit - Open-source WebRTC media server that realtime voice and video agent apps connect through.
• LiveKit Agents - Framework for realtime voice and video AI agents, with plugins for speech, LLM and TTS services.
• Pipecat - Python framework for real-time voice and multimodal conversational agents with pluggable services.
• Speech to Speech - Modular voice agent pipeline chaining voice activity detection, speech recognition, LLM, and synthesis.
• VoiceMem - Memory infrastructure for voice agents that keeps conversation context across sessions.

Rising

New projects under 500 stars that are worth watching. They move to their category at 500 stars.

• Agentrove - Self-hosted workspace to run and orchestrate several AI coding agent CLIs in one place. (Coding Assistants)
• Basjoo - Self-hosted AI customer support platform with per-knowledge-base RAG on Qdrant. (RAG and Vector Databases)
• flowcat - Self-hosted Rust runtime for real-time voice agents, shipped as a single binary. (Voice Agents)
• franken_ocr - CPU-only OCR engine written in Rust that runs the Unlimited-OCR vision-language model. (MIT with OpenAI and Anthropic rider) (Document AI and OCR)
• fxhoudinimcp - MCP server that exposes SideFX Houdini scene control to AI agents. (MCP Servers and Gateways)
• Guaardvark - Self-hosted AI studio on your own GPU for chatting with your files and running agents. (Agents and Automation)
• gyotaku - Offline Linux app that searches your screenshots by the text inside them. (Search and Knowledge)
• keirouter - Self-hostable AI gateway that proxies requests to multiple model providers. (Model Gateways and Routers)
• kensaku - Local search across documents, code and screenshots, indexed on your own machine. (Search and Knowledge)
• Kindly Web Search MCP Server - MCP server for web search and robust page content retrieval in AI coding tools. (MCP Servers and Gateways)
• llamastash - Terminal-native orchestrator that launches and manages local LLM engines and models with low overhead. (LLM Runtimes and Inference)
• llmserve - Terminal UI that picks a local model and backend, then starts a server for it. (LLM Runtimes and Inference)
• Lumiverse - Self-hosted AI chat frontend with an extension ecosystem and a focus on performance. (Lumiverse Community License 2.2, not OSI) (Chat Interfaces)
• Mira - Self-hosted AI code reviewer with indexed pull request reviews and walkthroughs. (Coding Assistants)
• mlxtop - Terminal monitor for local LLMs showing memory, GPU use and tokens per second. (Observability and Evals)
• Myna - Lightweight speech-to-text desktop application for Ubuntu. (Speech and Audio)
• Nyro - Self-hosted AI gateway that lets coding CLIs and SDKs call any model provider. (Model Gateways and Routers)
• openasr - Local-first speech-to-text command line tool that runs without cloud services or telemetry. (Speech and Audio)
• PMetal - Apple Silicon framework for local LLM inference and LoRA or QLoRA fine-tuning. (Fine-Tuning and Training)
• ScribeCrate - Self-hosted Whisper speech-to-text server powered by faster-whisper. (Speech and Audio)
• shodh-memory - Local, LLM-free memory engine for AI agents, packaged as one offline Rust binary. (Agents and Automation)
• thunderbird-mcp - MCP server that gives AI assistants access to Thunderbird email, contacts and calendar. (MCP Servers and Gateways)
• VoicEra - Voice AI building block with telephony integration and real-time speech processing. (Voice Agents)
• YoloRouter - Self-hosted OpenAI-compatible LLM gateway with provider failover and key rotation. (Model Gateways and Routers)
• zx-bench - Local LLM benchmarking platform with versioned datasets and deterministic scoring. (Observability and Evals)

Source-Available

Widely used and self-hostable, but the license adds terms beyond a standard open-source license. Check them before commercial use.

• Activepieces - Self-hostable automation platform that adds AI agents and MCP integrations to workflow building. (MIT core, ee directory under separate terms)
• Agenta - Open-source LLMOps platform for prompt management, evaluation and observability. (MIT core, ee/ directories under their own license)
• AutoGPT - Platform for building, running and monitoring autonomous AI agents with a visual workflow builder. (Polyform Shield for platform, MIT elsewhere)
• Browserless - Headless Chrome service that agents and scripts drive over an API for browser automation at scale. (MongoDB SSPL or commercial license)
• Crush - Terminal-based AI coding agent that works with multiple LLM providers and project context. (FSL-1.1-MIT (converts to MIT after two years))
• Dify - Platform for building agentic workflows and LLM apps with a visual editor, RAG pipeline and model management. (Modified Apache-2.0 with extra terms)
• FastGPT - Knowledge base and workflow platform for building RAG chatbots over your own documents. (Apache-2.0 with additional conditions)
• Fish Speech - Text-to-speech and voice cloning model and server for multilingual speech synthesis. (Fish Audio Community License 1.0)
• Foundry Local - Microsoft on-device runtime with a CLI and local API for running open models on your own hardware. (SDK is MIT; CLI under Microsoft license terms)
• HunyuanVideo - Large open video generation model from Tencent for text-to-video creation on your own GPUs. (Tencent Hunyuan Community License, excludes EU/UK/KR)
• IndexTTS - Zero-shot text-to-speech model that clones a voice from a short reference audio clip. (bilibili Model Use License Agreement)
• Langfuse - LLM engineering platform for tracing, prompt management and evals, self-hostable with Docker. (MIT core, ee/ directories under their own license)
• LiteLLM - Python SDK and proxy server that calls 100+ LLM APIs in the OpenAI format. (MIT core, enterprise/ directory under its own license)
• LobeChat - Modern chat UI for many LLM providers with plugins, knowledge bases and multimodal input. (LobeHub Community License)
• LTX-2 - Open video generation model and pipeline for text-to-video and image-to-video on your own GPUs. (LTX Community License Agreement)
• Meilisearch - Search engine with typo tolerance and hybrid and vector search, run as a single self-hosted binary. (MIT core, EE parts under Business Source License 1.1)
• MinerU - Converts PDFs into Markdown and JSON with layout, table and formula recognition, on local GPUs or CPUs. (Apache-2.0 plus commercial-use terms in LICENSE)
• n8n - Workflow automation platform with code and no-code nodes, including AI agent and LLM building blocks. (Sustainable Use License)
• Open WebUI - Self-hosted web interface for chatting with local and cloud LLMs, with RAG, tools and user management. (Open WebUI License)
• Phoenix - Tracing, evaluation and prompt experimentation tool for LLM apps, built on OpenTelemetry. (Elastic License 2.0 (ELv2))
• Qwen-Image-2.1 - Open image generation model from the Qwen team, distributed under the Qwen research license. (Qwen Research License Agreement)
• TEN Framework - Framework for real-time multimodal voice agents with modules for ASR, LLM, TTS and avatars. (Apache-2.0 with additional conditions)
• Wan2GP - Low-VRAM app for running Wan and other open video generation models on consumer GPUs. (WanGP Community License 2.0)
• Weaviate - Vector database with hybrid search and built-in embedding modules, runnable on your own servers. (BSD-3 core, wl/ directory under Weaviate License)

Contributing

See CONTRIBUTING.md. New projects go in data/projects.json; this file is generated.