Project directory

Projects

Open-source AI projects tracked by the leaderboard, ordered by GitHub stars.

TypeScript 8 stars

steve-aldrin/vecstore-js

A simple, fast, and pluggable vector storage for the browser

### Overview: VecStore-js is a fast, pluggable vector storage solution for the browser, perfect for offline-first AI features, browser extensions, and web apps that prioritize data privacy. ### Key features: - 🧠 Local Semantic Search: No server calls needed, all processing happens in the browser. - 🚀 High Performance: HNSW (Hierarchical Navigable Small World) for fast, approximate nearest-neighbor search, implemented in WebAssembly. - 🔌 Pluggable Components: Easily switch between search algorithms (HNSW, Cosine Similarity) and embedding models. - 🔒 Privacy-First: User data never leaves the browser. - Offline-First: Caches models and data for use without an internet connection. - Lightweight: Small footprint, designed for the client-side. ### Tech stack: TypeScript ### Why it matters: VecStore-js simplifies local semantic search, making it ideal for applications requiring privacy and offline-first functionality.

Python 8 stars

steveimm/comfyui-lance-aio

Run ByteDance Lance-3B (unified image/video generation, editing, and understanding) in ComfyUI, with dynamic VRAM streaming for low-VRAM GPUs

### Overview: This repository implements ComfyUI custom nodes to run Lance-3B model using ComfyUI, focusing on tasks such as text-to-image, text-to-video, image-text to video, image editing, video editing, and image understanding. It supports low-VRAM GPUs and is compatible with limited VRAM machines. ### Key Features: - Text-to-Image - Text-to-Video - Image-Text to Video - Image Editing - Video Editing - Image Understanding - Video Understanding ### Tech Stack: Python ### Why It Matters: This implementation allows for efficient use of low-VRAM GPUs, making it suitable for a wide range of devices with limited computational resources. It leverages ComfyUI's capabilities to provide a seamless user experience while maintaining high performance.

Python 8 stars

streamweights/streamweights

Build a small model for your task, on hardware you control. Bring labeled examples. Fine-tune locally, compare against simple baselines, and export to GGUF or safetensors. Pause and resume supported training runs across Mac and Linux.

Overview: Streamweights is a tool designed for developers that allows them to build small models for specific tasks on hardware they control. This approach enables them to fine-tune models locally and compare their performance against simple baselines, ensuring that the model is worth building. Key features: - Training a small model from labeled examples - Measuring the model's performance against simpler approaches - Exporting the model in GGUF or safetensors format - Support for Mac and Linux platforms - Ability to pause and resume training runs - Data stays local by default, with remote storage used when configured Tech stack: Python Why it matters: Streamweights simplifies the process of building and evaluating models, making it accessible for developers without requiring extensive machine learning expertise. It allows for fine-tuning models locally, ensuring that the model's performance is accurate and relevant to the task at hand. Repository: streamweights/streamweights GitHub stars: 8 GitHub description: Streamweights is a tool designed for developers that allows them to build small models for specific tasks on hardware they control. This approach enables them to fine-tune models locally and compare their performance against simple baselines, ensuring that the model is worth building.

TypeScript 8 stars

sujal12344/AI-Companion-Builder

AI-Companion is a cool software that lets you create your own custom AI models of people you admire, like actors or celebrities. It's a tool to make personalized artificial intelligence companions based on your favorite individuals.

Overview: AI-Companion is a software tool that allows users to create their own AI companions based on their favorite celebrities or actors. It provides a dynamic SaaS application built on Next.js 14, TypeScript, and other advanced technologies. The platform offers features like engaging with pre-existing AI models, creating custom models, managing models, and searching for existing models. Key features: - Engage with pre-existing AI models of celebrities - Create and customize your own AI models - Manage and update custom AI models - Search for existing AI models by name or category Tech stack: Next.js, Node.js, TypeScript, TailwindCSS, Stripe, Zod, Prisma, React Hook Form, Shadcn, zustand, OpenAI, Pinecone-database, Upstash, LangChain, Replicate Why it matters: AI-Companion empowers users to create personalized AI companions that can provide insights and engage them in meaningful conversations, making the experience of interacting with AI more personalized and engaging.

Rust 8 stars

sunormesky-max/epicode

Epicode - AI Memory Operating System. Spatial AI memory with knowledge graph, MCP integration, and auto-defense.

### Overview: Epicode is an open-source AI memory service designed for AI agents, providing spatial memory, semantic search, knowledge graph, temporal validity, MCP integration, and more. ### Key Features: - **Spatial Placement** - Memories are placed as tetrahedrons for clustering and visualization. - **Semantic Search** - Hybrid search using BM25 and HNSW for natural-language retrieval. - **Knowledge Graph** - Automatic relationship extraction and dynamic graph updates. - **Temporal Validity** - Supports `valid_from` / `valid_until` windows and freshness scoring. - **MCP Integration** - Discover the current tool catalog with MCP `tools/list`. - **L0 Active Inference** - Evidence-backed drive signals with REST/MCP inbox and acknowledgement flows. - **SMRP Protocol** - Transport-independent structured memory responses with tier, retrieval provenance, and optional topology/placement metadata. - **Multi-tenant Cloud** - User management, quotas, invite codes, and cloud deployment. ### Tech Stack: Rust ### Why it Matters: Epicode provides a robust and scalable solution for AI agents, enhancing their ability to store, retrieve, and manage spatial and temporal memories, making them more effective in various applications. ### Repository: sunormesky-max/epicode ### GitHub Stars: 8

Dockerfile 8 stars

suriya-mca/Docker_Compose_Files

In this repo, we have different docker-compose files for open-source and popular software and frameworks

Overview: This repository contains Docker Compose files for various open-source and popular software and frameworks, designed to set up these software or frameworks in a containerized environment. Key features: - Docker Compose files for SQL databases (SQLite3, PostgreSQL, CockroachDB, MySQL, MSSQL, Mariadb, TimescaleDB, EdgeDB) - Docker Compose files for NoSQL databases (Redis, MongoDB, KeyDB, Cassandra, CouchDB, Couchbase, Neo4j, Memcached, Valkey) - Docker Compose files for Vector databases (Qdrant, ChromaDB, PostgresVector, Weaviate) - Docker Compose files for Analytics databases (ClickHouse, Apache Pinot, Apache Kylin, Apache Druid) - Docker Compose files for DB clients (Dbeaver, Dbgate, Pgadmin4, Phpmyadmin, Beekeeper Studio) - Docker Compose files for Backend frameworks (FastAPI, Django, Spring Boot) - Docker Compose files for CRM/ERP (Odoo, ERPNext) - Docker Compose files for CMS (WordPress, Directus, Ghost, Grav, Strapi) - Docker Compose files for Docker clients (Portainer, Dokku) - Docker Compose files for Backend as a Service (Pocketbase, Appwrite) - Docker Compose files for API Platform (Hoppscotch) Tech stack: Dockerfile Why it matters: These Docker Compose files provide a standardized way to set up and manage open-source software and frameworks in a containerized environment, making it easier to deploy

Python 8 stars

thatgirlfrommoon/Doc-Chat

A Chatbot buddy for any websites

Overview: Doc-Chat is a versatile chatbot designed to extract insights and knowledge from any website or document, making it an invaluable tool for web documentation, PDF books, and recent publications. Key features: - Crawling and extracting information from websites and documents - Vector database storage for persistent knowledge - Easy installation and setup with Python and LangChain libraries Tech stack: Python, LangChain, uv, vector_db Why it matters: Doc-Chat streamlines the process of extracting knowledge from various sources, making it a valuable tool for researchers, students, and anyone looking to enhance their understanding of new or unfamiliar topics.

C# 8 stars

tryAGI/Pinecone

C# SDK for the Pinecone API -- vector database with similarity search, index management, and inference

Overview: Pinecone is a C# SDK for interacting with the Pinecone vector database, providing similarity search, index management, and inference capabilities. Key features: - Fully generated C# SDK based on the official Pinecone OpenAPI specification. - Supports updates to the database on the same day. - Automatically updated and supported without breaking changes. - Utilizes modern .NET features such as nullability, trimming, and NativeAOT. - Compatible with .NET Framework and .NET Standard 2.0. Tech stack: C# Why it matters: Pinecone simplifies working with vector databases, enabling developers to easily integrate vector search capabilities into their applications.

Python 8 stars

tryjmmj1-at/flux-2-flex-prompts-de

FLUX.2 Flex prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`flux-2-flex`) and the cost per image.

### Overview: FLUX.2 Flex Prompts is a collection of 8 tested image prompt recipes designed for developers. Each recipe includes a specific subject, light direction, framing ratio, and setting, along with the cost of the call on the APIMart route. ### Key Features: - **8 Recipes**: Each recipe is designed to be used as-is or modified for different purposes. - **Example Renders**: Detailed renders of each recipe, including the cost on the APIMart route. - **Model ID**: `flux-2-flex`, ensuring compatibility with the OpenAI-compatible endpoint. - **Effective Price**: $0.04 per image, with a total estimated cost of $0.32 for the set. ### Tech Stack: Python ### Why It Matters: Teams can leverage FLUX.2 Flex (`flux-2-flex`) to quickly generate high-quality images without the need for expensive subscription plans. The API key allows for easy integration and scaling, making it ideal for developers and creatives looking to experiment with different prompts and models.

Python 8 stars

tw31uj37/doubao-seedance-2-5-api-ja

シードダンス / シーダンス — Seedance 2.5 API (`seedance-2.5`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint.

Overview: Seedance 2.5 API is a model id, pricing, and parameters API for OpenAI-compatible APIMart endpoint, providing comprehensive access to 300+ models including video, image, and language models. Key features: - Supports per-second billing by resolution - Provides a single API key for catalog access - Offers a flexible pricing model with no subscription or prepaid plans - Reports complete tasks with their exact cost and failed tasks with a refund Tech stack: Python Why it matters: Seedance 2.5 API simplifies the process of accessing high-quality video and language models, making it easier and more cost-effective for developers to integrate with APIMart's OpenAI-compatible endpoint. Repository: tw31uj37/doubao-seedance-2-5-api-ja GitHub stars: 8 GitHub description: Seedance 2.5 API (`seedance-2.5`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint.

Python 8 stars

uncoalesced/Peridot

Peridot v1.6.0 [STABLE] | Sovereign AI Kernel. Offline, GPU accelerated Qwen framework: a 27B model fully on an 8GB GPU at 23.6 tok/s. Sandboxed skills & plugins, model tool use, opt in web search, MCP bridge for private data jobs from cloud agents, ZAT-SCS preemption, RAG. Zero data exfiltration. Engineered by uncoalesced.

### Overview: Peridot v1.6.0 is a sovereign local AI kernel designed for offline inference, with hardware-aware GPU arbitration and predictive context preparation. Every inference cycle runs on hardware the operator owns, ensuring no cloud dependency, telemetry, or remote trust assumptions. The default model is now **Qwen3.8-27B (UD-IQ1_S)**, fully GPU-resident on an 8GB card, offering 23.6 tokens/sec decode and ~500 tokens/sec prefill, up from 5.3 and 34 on the old 14B default. This is a beta release, but the 27B install path has been validated on one machine (RTX 5050 Laptop, 8GB). The new v1.6.0 (stable) version makes Peridot agentic, preserving user sovereignty. ### Key features: - **GPU-accelerated inference**: Runs on hardware owned by the operator, ensuring no cloud dependency. - **Hardware-aware GPU arbitration**: Predictive context preparation for optimal performance. - **Native Linux support**: Added support for Debian 12, Ubuntu 22.04+, and Arch. - **Zero telemetry**: No telemetry or remote trust assumptions. - **Native CUDA kernels**: Fully GPU-resident on an 8GB card, offering 23.6 tokens/sec decode and ~500 tokens/sec prefill. - **Native 27B model**: Qwen3.8-27B (UD-IQ1_S

CSS 8 stars

vishwaswarup/BiteRate

Biterate – Restaurant Review Sentiment Analyzer An NLP-powered sentiment analysis platform using emoji-aware preprocessing, TF-IDF (1–3 gram) features, and model benchmarking across Logistic Regression, Naive Bayes, Linear SVC, XGBoost, and a Soft Voting Ensemble, achieving up to 88% accuracy and 0.87 F1 Score

### Overview: BiteRate is an NLP-powered sentiment analysis platform designed to classify restaurant reviews into positive, neutral, or negative categories. Leveraging advanced preprocessing techniques and machine learning models, it achieves high accuracy and robust performance. ### Key Features: - **Emoji-aware Preprocessing:** Handles contractions, lemmatization, negation marking, and emojis. - **TF-IDF Features:** Incorporates 1-2-gram word and 2-5-gram character features for comprehensive analysis. - **Model Benchmarking:** Uses Logistic Regression, Naive Bayes, Linear SVC, XGBoost, and a Soft Voting Ensemble for accurate classification. - **High Accuracy:** Achieves up to 88% accuracy and 0.87 F1 Score, demonstrating its effectiveness across various datasets. ### Tech Stack: Python, Flask, scikit-learn, NLTK, TF-IDF, logistic regression, Naive Bayes, Linear SVC, XGBoost, Soft Voting Ensemble. ### Why It Matters: BiteRate provides a reliable sentiment analysis tool for restaurant reviews, helping businesses understand customer sentiment and improve their offerings. Its robust model and user-friendly interface make it an essential tool for restaurant owners and marketers. ### Repository: vishwaswarup/BiteRate GitHub Stars: 8 GitHub Description: An NLP-powered sentiment analysis platform using emoji-aware preprocessing, TF-IDF features, and model benchmarking across Logistic Regression, Naive Bayes, Linear SVC, XGBoost, and a Soft Voting Ensemble, achieving up

Python 8 stars

visresearch/HFPrune

Overview: This repository contains the official implementation of the paper "HIGH-FIDELITY PRUNING FOR LARGE LANGUAGE MODELS". It provides a novel pruning method using information entropy for importance evaluation, aiming to achieve superior performance compared to existing approaches. Key features: - Pruning a LLaMA model using Taylor-based importance estimation - Fine-tuning with LoRA (Low-Rank Adaptation) for efficiency - Supports full fine-tuning and LoRA fine-tuning - Evaluates on various benchmarks, including Hellaswag, PIQA, ARC Challenge, and more Tech stack: Python Why it matters: This method surpasses traditional pruning techniques and achieves consistent improvements across various benchmarks, making it a significant advancement in the field of large language model pruning. Repository: visresearch/HFPrune GitHub stars: 8 GitHub description: No description provided. README excerpt: No description provided.

Jupyter Notebook 8 stars

vkhamesi/proteins

🧬 Fine-Tuning Large Language and Protein Models on a single T4 GPU via Distillation, Quantization and Low-Rank Adaptation to run inference on proteins functions.

### Overview: Fine-Tuning Large Language and Protein Models on a single T4 GPU via Distillation, Quantization and Low-Rank Adaptation to run inference on proteins functions. ### Key features: - Fine-tuning of large language models (BioBERT) and protein models (ProtBERT) for improved protein function prediction. - Utilization of mixed precision and quantization techniques for efficient model training on a single T4 GPU. - Distillation of ProtBERT to enhance the performance of BioBERT for protein function prediction. - Low-rank adaptation of models to reduce computational complexity and improve inference speed. ### Tech stack: Jupyter Notebook ### Why it matters: This approach allows for the efficient training and inference of large models on a single GPU, enabling faster and more accurate predictions of protein functions.

Python 8 stars

w4ez8p7hr0h5o/wan-2-7-image-api-es

Wan 2.7 Image API (`wan2.7-image`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint.

Overview: Wan 2.7 Image API provides access to a wide range of image models through an OpenAI-compatible endpoint, including Wan 2.7 Image, with pricing details and API parameters included. Key features: - Wan 2.7 Image (`wan2.7-image`) model - Pricing details - API parameters for prompt, model, size, resolution, and more Tech stack: Python Why it matters: It offers a single endpoint for OpenAI-compatible image generation, simplifying integration and reducing costs compared to multiple endpoints. Repository: w4ez8p7hr0h5o/wan-2-7-image-api-es Primary language: Python GitHub stars: 8 GitHub description: Wan 2.7 Image (`wan2.7-image`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint.

Python 8 stars

warwick-bit/llm-accuracy

Evidence-first accuracy hygiene for Claude.

Overview: LLM Accuracy is a tool designed to enhance the accuracy of Claude's responses by prompting it to clarify ambiguous questions, align with evidence, and address conflicting sources, ensuring more consistent and reliable answers. Key features: - Clarify ambiguous questions - Align answers with evidence - Address conflicting sources - Apply reviewed definitions and declared source routes Tech stack: Python Why it matters: By improving the consistency and reliability of Claude's responses, LLM Accuracy helps users rely on more predictable and trustworthy answers, reducing the risk of incorrect conclusions and ensuring better decision-making in various applications.

Python 8 stars

wbott/rag-movie-rec

Production-ready RAG movie recommender with multi-agent architecture, OpenAI embeddings, and conversational web UI. Ask naturally about movies - get intelligent recommendations instantly!

Overview: This project is a production-ready RAG movie recommender system that leverages multi-agent architecture, OpenAI embeddings, and a conversational web UI. It allows users to ask natural questions about movies, receiving intelligent recommendations instantly. Key features: - 🎬 Conversational Movie Search - 🎯 Smart Recommendations - 🚀 Instant Movie Info - 🤖 Multi-Agent Architecture - 🔍 Vector-Powered Search Tech stack: Python Why it matters: This system addresses the shortcomings of traditional movie search by understanding context, meaning, and relationships between movies, providing more accurate and personalized recommendations. Repository: wbott/rag-movie-rec GitHub stars: 8 GitHub description: Production-ready RAG movie recommender with multi-agent architecture, OpenAI embeddings, and conversational web UI. Ask naturally about movies - get intelligent recommendations instantly!

Python 8 stars

wuxuedaifu/vllm-chatterbox-stream

OpenAI-compatible multilingual TTS server — Chatterbox on vLLM with real-time PCM audio streaming, low time-to-first-byte (~0.7 s), voice cloning, and 23 languages.

Overview: This project is an OpenAI-compatible multilingual TTS server built on vLLM, featuring real-time PCM audio streaming, low time-to-first-byte (~0.7s), voice cloning, and support for 23 languages. Key features: - Real-time PCM audio streaming - Low time-to-first-byte (~0.7s) - Voice cloning - Support for 23 languages Tech stack: Python Why it matters: This project leverages vLLM's capabilities to provide high-quality multilingual TTS with real-time audio streaming, making it ideal for applications requiring seamless, low-latency audio delivery. Repository: wuxuedaifu/vllm-chatterbox-stream Primary language: Python GitHub stars: 8 GitHub description: OpenAI-compatible multilingual TTS server — Chatterbox on vLLM with real-time PCM audio streaming, low time-to-first-byte (~0.7 s), voice cloning, and 23 languages. README excerpt:

Python 8 stars

x69ynmals73l4z/z-image-prompts-ko

Z Image Turbo prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`z-image-turbo`) and the cost per image.

### Overview: This repository contains 8 tested image prompt recipes, including example renders, aspect ratios, the model ID (`z-image-turbo`), and the cost per image on the APIMart route. It provides a concise guide for developers to create effective and cost-efficient image prompts. ### Key Features: - **Prompt Recipes:** Eight unique image prompts with detailed instructions. - **Example Renders:** Visual examples of each recipe. - **Aspect Ratios:** Specific ratios for optimal image quality. - **Model ID:** `z-image-turbo` for compatibility with OpenAI models. - **Cost Calculation:** Estimated cost per image on the APIMart route. - **Prompt Structure:** A cheat-sheet for structuring prompts effectively. ### Tech Stack: - Python ### Why It Matters: - **Efficiency:** Simplifies the process of creating image prompts by providing a structured approach. - **Cost-Effectiveness:** Offers cost estimates and efficient usage of the model, making it a valuable tool for developers. - **Compliance:** Ensures compatibility with OpenAI models, making it a reliable resource for developers working with image prompts.

Python 8 stars

xltzsoft/deepseek-v4-sm89

面向 8×RTX 4090（SM89）的 DeepSeek-V4-Flash-0731 MXFP4 SGLang 实验分支

Overview: This is a DeepSeek-V4-Flash-0731 MXFP4 experiment branch specifically designed for NVIDIA RTX 4090 (SM89) with 8×RTX 4090 24 GB. The repository contains only the inference framework source code, not including model weights, JIT cache, benchmark requests, access credentials, or server-specific configurations. Key features: - MXFP4 routed experts - Target-only mode with decode full CUDA Graph - Kernel optimizations for SM89 architecture - Performance benchmarks for 8×RTX 4090 24 GB with TP=8 Tech stack: Python Why it matters: This branch is designed for a single machine with 8 NVIDIA RTX 4090 GPUs, targeting the Ada SM89 architecture for inference. It includes optimizations for the SM89 GPU, including kernel routing for MXFP4 Marlin MoE, sparse decode, and fallbacks for certain components. The performance benchmarks show an efficiency of 87.5% for the Scheduler output, 85.5% for the decode window, and 72.7% for the end-to-end aggregation, indicating significant improvements over the current production version. GitHub stars: 8