uni‑api is a config‑only server that presents a single OpenAI‑compatible API while routing calls to many LLM providers (OpenAI, Anthropic, Gemini, Vertex, Azure, AWS, etc.). It supports weighted and round‑robin load balancing, automatic retries, per‑model timeouts, rate limits, tool‑call passthrough, content moderation, and fine‑grained request overrides—all defined in an `api.yaml` file or via a `CONFIG_URL`. Deployable as a Docker container (one‑click Fugue button provided).

A set of Flutter plugins that enable mobile apps to use Google's standalone ML Kit for on-device vision, natural language, and generative AI features.

A method for improving LLM generalization during Supervised Fine-Tuning by dynamically rescaling token loss based on predicted probabilities, particularly effective for reasoning tasks.

A pure Go Japanese morphological analyzer that tokenizes text and analyzes parts of speech with embedded dictionaries for easy deployment.

A cross-platform multimedia processing framework by ByteDance that enables high-performance video transcoding, editing, and AI inference with strong GPU acceleration.

VectorVein is a cross‑platform, no‑code desktop app for building AI‑driven automation workflows. It lets users visually connect LLM calls, embeddings, speech‑to‑text, and other nodes, supports any OpenAI‑compatible endpoint (OpenAI, Ollama, LM‑Studio, etc.), provides a local FastAPI REST API for programmatic execution, and ships as a single executable built with pywebview.

A serverless Python runtime and client library for deploying and scaling ML models and pipelines in the cloud without managing infrastructure.

A PyTorch-based supervised semantic segmentation toolbox providing a unified framework and extensive model zoo for training and testing segmentation algorithms.

A JavaScript library for AI-powered image enhancement, allowing developers to increase resolution, denoise, and deblur images in the browser or Node.js.

The backend API for the Hugging Face dataset viewer, allowing users to browse, filter, and inspect datasets on the Hub through pre-computed data.

A neural network compiler for AI accelerators that optimizes and deploys models from TFLite, Caffe, and ONNX formats onto specialized hardware like the K210, K510, and K230.

A browser-based web application for recognizing text from images, proofreading OCR data with high precision, and creating fully digitized, ebook-style documents.

A high-performance, cross-platform inference engine for running LLMs and VLMs locally on desktop, mobile, and edge devices.

An embeddable, in-memory document database written in C++ and Go that provides a high-performance alternative to Elasticsearch with support for vector and full-text search.

A Python framework for building memory-first AI agents that use persistent episodic, semantic, and procedural memory to recall experience and learn new skills over time.

A machine learning tool that automatically ranks strings extracted from binaries to help malware analysts quickly identify the most relevant information.

Official Python SDK for Mistral AI’s cloud APIs (chat, embeddings, audio, files, agents, etc.). Install via pip/uv/poetry, set `MISTRAL_API_KEY`, then call methods like `client.chat.complete(...)` synchronously or with `asyncio`. Includes Azure and GCP wrappers and support for streaming, pagination, retries, and more.

A framework for the Cline VS Code extension that manages large-scale project context and dependencies using a recursive, file-based tracking system and local LLM-assisted analysis.

GPU‑Jupyter provides a Docker‑based, GPU‑enabled JupyterLab environment pre‑installed with Python, PyTorch, TensorFlow, Julia, and R. It offers versioned images, easy customization, and Docker‑Compose support, enabling reproducible deep‑learning experiments on any machine with an NVIDIA GPU.

TypeAgent is a Microsoft open‑source sample that shows how to build a single personal AI assistant using large language models, structured prompting, and a novel “Structured RAG” memory system. It includes an Electron shell, a dispatcher that routes natural‑language requests to typed agents, a memory layer that stores conversation facts as logical entities, and a cache to reduce LLM calls. The repo provides many example agents (calendar, email, browser, etc.) and an SDK for adding custom agents. It is early‑stage sample code, tested with Azure OpenAI in English, and not intended for production without further validation.

A unified toolkit for training and deploying billion-scale neural retrievers across different languages and modalities using LLMs.

A compact, C-based machine learning runtime with a Python API that provides full control over tensor execution and memory for research and low-level optimization.

A tool that converts Jupyter Notebooks into production-ready Kubeflow Pipelines by tagging cells, removing the need to rewrite code using the KFP SDK.

pnnx is an open‑source tool that optimises PyTorch models and exports them to a lightweight, dependency‑free format (PNNX) and to ncnn/ONNX‑zero files, enabling fast inference on edge devices.

An AI-powered document processing engine for ETL and RAG that transforms unstructured documents into high-quality data chunks for vector databases.

LitData is a Lightning AI library that speeds up PyTorch training by providing fast, cloud‑native data pipelines. It offers two main modes: (1) `StreamingRawDataset` streams raw files directly from S3/GCS/Azure/HF Hub with async, batched downloads; (2) `ld.optimize` converts a dataset into LitData’s own chunked binary format, enabling up‑to‑20× faster streaming, shuffling, and resumable epochs. The library also includes a parallel `map` primitive for preprocessing (e.g., image resizing, embedding creation) and full support for many modalities (images, audio, video, tensors, graphs, etc.). Integration points include PyTorch Lightning, Hugging Face datasets, and Lightning Cloud, making it a practical tool for large‑scale ML projects where I/O is the bottleneck.

PaperForge Research OS v3 is a Python‑based tool that unifies AI‑assisted paper writing, experiment orchestration, artifact tracking, and verifiable publishing. It stores every claim‑to‑evidence link in a SQLite “Scientific Memory”, enforces three execution profiles (writing‑only, research, full), and supports local, Docker, SSH, Slurm, Kubernetes and cloud compute back‑ends. The CLI (`paperforge …`) and a local web UI let users run workflows, approve proposals, compile LaTeX with built‑in templates (generic, CVPR, IEEE, Elsevier), and release a deterministic source bundle after secret‑scanning. Designed for reproducible academic work, it is open‑source under a non‑commercial license.

A single-stage 3D scene reconstruction tool that creates 3D point clouds and camera poses from sparse RGB views in about 2 seconds without requiring pre-known camera poses.

A collection of Topological Data Analysis (TDA) Python libraries designed to make topological data tools accessible to non-topologists.

Julia bindings for Enzyme, a high-performance automatic differentiator that operates on LLVM to provide efficient gradients for optimized code.

An open-source framework for globally distributed training of large models that reduces communication overhead using distributed weight averaging.

A comprehensive evaluation library for NLP and LLM models to test for robustness, bias, fairness, and safety.

An open-source workspace for AI agents that provides persistent, recoverable sessions and a unified runtime for software and knowledge work.

SITS is an open‑source R package (with a Python wrapper) for building, regularising, and classifying satellite‑image time‑series data cubes. It connects to many public EO collections, offers a suite of ML/DL models (including GPU‑accelerated CNNs and attention encoders), and provides end‑to‑end tools for training, classification, smoothing, uncertainty estimation, and active learning.

A Java thread dump and heap analysis tool that provides a GUI and an MCP server for diagnosing deadlocks, performance issues, and virtual thread pinning.

A Python package for detecting and segmenting grains in images using a hybrid U-Net and SAM 2.1 approach, designed for geomorphology and sedimentary geology research.

A pure Python machine learning library built from scratch for educational purposes, providing transparent implementations of core ML algorithms and an autograd engine.

A PyTorch-based module for implementing and testing Echo State Networks and Reservoir Computing, designed for research purposes and integration into deep architectures.

A NASA-developed C++ library for general-purpose image processing and computer vision, providing tools for camera models, cartographic projections, and mosaic compositing.

A framework for scaling reasoning-capable models through a self-improvement loop of synthetic data generation and automated verification.

A biomolecular modeling library used for the computational analysis and design of protein structures, enzymes, and biological macromolecules.

A set of plugins for radare2 that integrates LLMs to assist in reverse engineering tasks such as function explanation, vulnerability discovery, and augmented decompilation.

A Clojure library for fast tensor and neural network computations that leverages optimized native libraries for CPU and GPU acceleration.

An AI-native presentation platform that uses a JSON-based intermediate language called MulmoScript to generate multi-modal content including videos, podcasts, and slide decks.

A C/C++ library that extends ONNX Runtime with custom operators to handle pre- and post-processing for vision, text, and NLP models within the ONNX graph.

A toolchain for deploying optimized AI algorithms on low-cost embedded devices and microcontrollers, supporting computer vision, anomaly detection, and scenario-specific tasks.

An agent execution engine that replaces simple ReAct loops with a Plan-Observe-Replan (POR) system for more reliable complex task execution.

VizSeq is a Python toolkit for visual, interactive analysis of text‑generation outputs (MT, summarisation, captioning, speech/video translation, etc.). It computes many standard and embedding‑based metrics in parallel and displays results in Jupyter notebooks or a built‑in web app, with optional Fairseq integration.