TurboFieldfare is a Swift + Metal runtime that streams the Gemma 4 26B‑A4B model from SSD, enabling inference on Apple‑silicon Macs with as little as 8 GB RAM. It ships a native macOS app, a command‑line tool, a Swift library, and a loop‑back OpenAI‑compatible server, all under Apache 2.0.

An open-source AI workbench for scientific research that automates the research loop, including literature review, coding, and running experiments.

An AI-powered video editor featuring layer-based editing and motion graphics tools to help creators produce professional effects without heavy software.

An open-source data annotation platform for building high-quality visual datasets for computer vision, supporting image, video, and 3D annotation.

Doop is an open‑source, real‑time design canvas where humans and AI agents (via Claude Code or a built‑in Doop Agent) edit HTML frames together. It offers live multiplayer features, AI‑driven design tasks, private sharing, and can be self‑hosted with a single Docker command or `bun run dev`. The AI backend works with Anthropic (free tier) or user‑provided OpenAI/ChatGPT keys.

AISBench Benchmark is an open‑source evaluation framework for LLMs, multimodal models, and AI agents. It supports both accuracy (using many public datasets) and performance (latency, throughput, cache) testing of locally‑run or service‑based models, with extensive recent additions such as Harbor‑agent benchmarking, prefix‑cache stress tests, and response‑anomaly detection.

Voxtype is an open‑source, MIT‑licensed Linux app that provides local, real‑time voice‑to‑text dictation. It runs fast CPU models (Cohere Transcribe 9‑11× realtime) and supports GPU/Intel NPU acceleration, nine interchangeable transcription engines, multilingual support, meeting‑mode transcription, and deep desktop integration (Wayland/X11, hotkey binding, auto‑pause media, waveform OSD). All processing stays on the machine unless the user opts for a remote server.

A hardware plugin that allows the vLLM inference engine to run on Huawei Ascend NPUs, supporting a variety of open-source LLMs and multimodal models.

A durable memory system for AI coding agents that indexes past sessions across different tools and allows them to recall decisions and rationale via Hugging Face datasets.

A knowledge base of skills, rules, and validation hooks that teach AI coding agents to generate correct, compile-ready UdonSharp code for VRChat world development.

A GPU-accelerated simulation platform for developing, testing, and training AI-powered robots in realistic virtual environments.

SAM 3 (Segment Anything with Concepts) is Meta’s 848 M‑parameter foundation model that lets you prompt an image or video with free‑form text (or visual exemplars) and receive masks, boxes, and scores for *all* matching objects. It combines a DETR‑style detector and a SAM 2‑style tracker, introduces a presence token for fine‑grained prompt discrimination, and is trained on >4 M auto‑annotated concepts. The repo provides installation steps, example notebooks, and a new SA‑CO benchmark (270 K concepts) for evaluation.

A visualizer for Claude Code and Codex sessions that transforms text transcripts into interactive, live flow graphs for monitoring and reviewing agentic workflows.

An open-source, self-hosted CRM with a built-in MCP server that allows AI agents to perform CRM operations and analysis via 37 specialized tools.

PaperJury is a Claude Code (or Codex) plugin that uses LLM agents to perform a structured, pre‑submission review of LaTeX/Markdown papers. It runs a “review → adjudication → edit → re‑check” loop, classifying each comment as safe‑to‑fix, author‑required, or invalid. Three modes (direct‑edit, review, auto) let users request specific changes, run a mock peer‑review, or let the system operate unattended with safety guards. The tool includes deterministic scripts for LaTeX compilation, compliance checks, and a ledger of decisions. Benchmarks on 12 papers show a 0.656 F1 issue‑detection score and a 4× drop in unsafe edits. Installation is via the Claude Code marketplace or a simple git clone; a full “dog‑food” sample with before/after PDFs is provided.

Rapid‑MLX is a native Apple‑silicon inference engine that runs LLMs, diffusion, audio and video models locally and exposes a drop‑in OpenAI/Anthropic HTTP API, letting any client (LangChain, Claude Code, Aider, etc.) run fully offline on an M‑series Mac.

A unified cross-platform desktop application for AI-assisted annotation of text, image, video, and multimodal data, integrating state-of-the-art deep learning models to automate labeling.

baoyu-design packages Claude Design as a local agent skill, enabling full‑featured UI/UX design (mockups, prototypes, decks, design‑system import, Figma .fig decoding, PPTX export, etc.) directly inside agents like Cursor, Claude Code, or Codex, with all artefacts stored as self‑contained HTML in your repo.

A pre-trained, lightweight Voice Activity Detector (VAD) that identifies speech in audio streams across 6,000+ languages with high accuracy and low latency.

An MCP server that gives AI coding agents access to Figma design data, allowing them to implement designs more accurately than using screenshots.

A professional multi-account manager for Google Gemini and Claude that automates account switching and provides a local API proxy to bypass quota limits.

A VS Code extension that integrates DeepSeek models into the Copilot Chat model picker, enabling DeepSeek's reasoning and vision capabilities while keeping Copilot's native agent and tool-calling features.

UFO is an AI agent framework that enables complex automation across multiple devices and operating systems, utilizing a DAG-based orchestration system to coordinate tasks between Windows, Linux, and Android.

Loop Engineering for VLAnything is an end‑to‑end Python toolkit that lets you record, merge, audit, enrich, and iteratively improve multimodal robot‑learning datasets (RGB, RGB‑D, and future visuo‑tactile) for vision‑language‑action models. It provides a conservative, side‑car‑first quality audit, plug‑in semantic evaluators, language‑annotation pipelines, and a closed‑loop feedback loop for policy improvement, all compatible with the LeRobot data format and publishable to Hugging Face.

A library that ensures LLMs return reliable, structured JSON by using Pydantic for validation, type safety, and automatic retries.

Ix is a system intelligence tool that parses codebases into a persistent graph of symbols and relationships, allowing humans and AI agents to query structure instead of searching text.

A compact, 3D-printable open-source humanoid robot designed as an affordable platform for learning and experimentation in robotics.

A collection of human-reviewed instruction sets (skills) for AI agents to ensure they produce idiomatic, secure, and high-performance Go code.

A local-first desktop workspace for running multiple AI coding agents and shells side-by-side with persistent sessions and shared MCP server management.

A terminal-based manager for AI coding agents that consolidates multiple CLI agents into a single interface with live status tracking and integrated diff reviews.

A distributed LLM serving system that pools GPUs and memory across multiple machines to run large models via an OpenAI-compatible API.

An immersive AI companionship assistant that combines LLMs with emotion recognition, desktop vision, and Galgame-style visuals to create a reactive digital character.

A skill that generates prompts to force AI agents into a rigorous loop of building and blind-criticizing against a real-world quality bar until the output wins.

A self-hosted platform for cloud development environments and AI coding agents that uses Terraform to define workspaces and provides centralized AI governance.

A utility to align session files and SQLite indices in Codex after switching AI providers, ensuring old chat sessions remain usable.

Kimi K3 is an open-weight, 2.8T-parameter native multimodal MoE model designed for long-horizon coding, deep research, and complex agentic knowledge work.

A free, local audio stem separation tool that splits songs into isolated tracks like vocals and drums using Demucs, providing a DAW-style mixer for local playback and export.

A framework for automatically searching and optimizing the code-based scaffolds (harnesses) surrounding a fixed AI model to improve task-specific performance without retraining.

An Obsidian plugin that transforms notes into a connected, queryable knowledge base using graph-based retrieval (PPR) instead of vector embeddings.

An LLM post-training framework for RL scaling that integrates Megatron and SGLang to provide high-performance training and flexible data generation.

zerostack is a Rust‑written terminal AI coding assistant that talks to many LLM providers, offers multiple built‑in prompts (code, plan, review, debug, etc.), enforces a granular permission system, supports session persistence, optional sandboxing, and extensible features like advisor models, hooks, and status‑signal sockets—all while staying lightweight (≈ 16 MiB RAM, 26 MB binary).

An AI-native API gateway based on Istio and Envoy that provides unified management, observability, and hosting for LLM APIs and MCP servers.

An open-source computer use agent framework that operates real GUIs via mouse and keyboard, achieving human-level performance on the OSWorld benchmark.

ComfyUI‑H3‑FaceRefine is a custom‑node pack for ComfyUI that fixes MiniMax H3’s poor rendering of small faces in video. It detects faces, crops them to a larger canvas, runs H3 on the crops, and stitches the refined faces back. Includes auto‑ and manual‑face‑selection modes, hard‑cut detection, per‑frame denoise, and ready‑to‑use example workflows.

A reusable agent skill that enables LLMs to build and maintain a structured wiki of synthesized knowledge, moving beyond simple RAG by compiling sources into durable, cross-linked markdown pages.

An autonomous security research framework that chains static and binary analysis with LLM-powered validation to discover, exploit, and patch vulnerabilities.

A modular library for using and training state-of-the-art diffusion models to generate images, audio, and 3D molecular structures.

A Model Context Protocol (MCP) server that connects Claude AI to Ableton Live, enabling prompt-driven music production, track manipulation, and autonomous song arrangement.