A coordination layer for AI coding agents that turns disconnected terminal sessions into a durable, coordinated team with a shared ledger and multi-runtime support.

DeepSWE is a benchmark for measuring coding agents on 113 long-horizon software engineering tasks from real-world open-source repositories.

Boost is a command wrapper that compresses noisy shell output into structured context for AI coding agents, reducing token costs without sacrificing task performance.

A Python framework for building reliable AI applications by enforcing input/output risk validation and ensuring LLMs generate structured data.

An automated research workflow that searches, recommends, and analyzes academic papers from arXiv and Semantic Scholar, integrating the results directly into Obsidian notes.

An open-source feature store for machine learning that manages the flow of data from offline training sets to low-latency online inference.

An environment and benchmark for building and testing autonomous agents that can control Android devices across a wide variety of real-world apps.

A collection of simplified reproductions of frontier LLM reinforcement learning algorithms, designed to lower the GPU and code complexity barriers for studying Agentic-RL.

A session cleaner and prompt injection tool for AI coding assistants (Codex, Claude Code, OpenCode) designed to bypass refusal responses during security testing and CTF competitions.

Kong is a cloud-native API and AI gateway that centralizes routing, security, and orchestration for microservices, LLMs, and Model Context Protocol (MCP) traffic.

A lightweight, dependency-free C/C++ tensor library for machine learning designed for high efficiency and cross-platform portability.

A large-scale evaluation framework for assessing AI agents' ability to perform real-world vulnerability analysis and Proof-of-Concept exploit generation.

Skales is a closed‑source, locally‑run AI desktop and mobile agent that can chat, automate multi‑step goals, edit code, generate design/media assets, manage calendars, and retain memory—all without sending data to the cloud. It installs with a single click on Windows/macOS/Linux/Android/iOS and supports >15 LLM providers or fully offline models.

A tool output proxy that intercepts command-line calls from AI agents to provide fabricated or modified responses for red-team testing and guardrail bypass evaluation.

A CLI tool and Python library for interacting with multiple LLMs, both via remote APIs and local installations, featuring built-in logging and structured data extraction.

A Go implementation of the Model Context Protocol (MCP) that enables LLM applications to integrate with external data sources and tools using a standardized interface.

An open-source orchestrator for AI coding agents that enables parallel task execution, remote runtimes, and a unified review-first workspace.

A local debugger for AI agents that provides real-time trace streaming and a self-healing evaluation loop for fixing agentic bugs.

An adapter that enables the Claude Agent SDK to be used with any client that follows the Agent Client Protocol (ACP).

A tool that extracts a complete design system from any live website, generating developer-ready tokens, Tailwind configs, and Figma variables.

A multimodal data-processing framework that provides 200+ modular operators to clean, deduplicate, and curate large-scale datasets for foundation models.

An inference engine that enables running Mixture-of-Experts models larger than a device's RAM by streaming experts from flash storage in real-time.

A unified AI SDK for Laravel that provides a consistent API to interact with providers like OpenAI, Anthropic, and Gemini for agents, images, and audio.

A knowledge graph builder for Databricks that transforms tables into materialized graphs using LLM-powered ontology design and mapping.

A tool and field guide designed to identify and remove generic, AI-generated design patterns (AI slop) from web projects.

An agent skill that audits codebases and generates detailed, self-contained implementation plans for cheaper AI models to execute.

A high-density knowledge library and toolset for AI coding agents to build professional Godot 4.7+ games using expert-level patterns and best practices.

A compact, human-readable encoding of the JSON data model designed to minimize token usage in LLM prompts while maintaining high retrieval accuracy.

A reference implementation of the Jacobian lens for decoding internal language model activations into vocabulary tokens to analyze internal representations.

Pentest Swarm AI is an open‑source Go tool that runs dozens of autonomous LLM‑driven agents in parallel, coordinated via a shared Postgres blackboard with pheromone‑based stigmergy. It supports any LLM provider (Claude, Together AI, OpenAI‑compatible, Ollama, etc.), bundles a recon/exploit toolchain (ProjectDiscovery libs, nmap), and offers both an interactive terminal UI and a web dashboard. The swarm can target real hosts or bundled vulnerable labs, enforces scope, and produces evidence‑backed reports. Core features are stable; the swarm scheduler is in alpha, with further adapters and a fine‑tuned model planned.

An Android application that acts as an MCP server, allowing AI models to remotely control a device via accessibility services and screenshot capture.

An AI-engineering foundation framework that provides an architecture-aware harness and modular business components to ensure AI-generated code remains consistent and architecturally sound.

An interactive visualization tool that runs a live GPT-2 model in the browser to help users learn how Transformer-based models predict text.

A self-hosted control plane for operating AI agents, allowing users to dispatch tasks, track spend, and coordinate multiple agent runtimes from a single dashboard.

A CLI tool that orchestrates AI coding agents into repeatable, YAML-defined development workflows with isolated worktrees and explicit review loops.

The official Python client and CLI for the Hugging Face Hub, enabling users to programmatically download, upload, and manage ML models, datasets, and demos.

An open lakehouse format for multimodal AI that provides high-performance vector search, random access, and native storage for images, video, and audio.

OpenGame is an open‑source, Node‑based framework that uses a specialised code LLM (GameCoder‑27B) to generate complete web games from a single natural‑language prompt. It provides a reusable *Game Skill* (template selection + automated debugging), a CLI (`opengame`) for headless generation, and an evaluation suite (OpenGame‑Bench) to measure playability. The system scaffolds projects with engines like Phaser, runs them in a sandbox, fixes integration errors, and outputs a ready‑to‑play HTML/JS game.

The Kubernetes MCP Server is a Go-based Model Context Protocol server that lets AI assistants and agents directly inspect and manage Kubernetes and OpenShift clusters through the Kubernetes API — not by wrapping kubectl or helm. It exposes tools for pods, namespaces, events, Helm releases, Tekton pipelines, KubeVirt VMs, and more, and can work across multiple clusters with no external command-line dependencies.

A modern cross-platform desktop toolset integrating image processing, audio/video editing, and local AI tools with GPU acceleration.

A collection of measurement-first agent skills that enable AI coding agents to audit and optimize web projects for performance, accessibility, and SEO using tools like Google Lighthouse.

A set of plugins for Claude Code and Codex that uses evidence-based learning science to provide interactive exercises, preventing skill decay during AI-assisted coding.

A tool for creating LLM fine-tuning datasets by converting unstructured documents into high-quality structured QA pairs and evaluation sets.

A governance standard and toolset for knowledge repositories maintained by LLM agents, ensuring deterministic work, resumability, and clear operator-agent boundaries.

A cross-device companion app and client for AI coding agents that lets you run sessions locally and control them remotely via mobile, web, or desktop.

vllm.cpp is a pure‑C++ inference engine that mirrors the vLLM Python server’s performance and features while being only ~66 MiB in size. It runs GGUF‑quantized LLMs (and multimodal models) on CPU, CUDA, Metal, Vulkan, etc., offering token‑for‑token identical output, continuous batching, speculative decoding, and an OpenAI‑compatible HTTP API—all without any Python runtime.

A Go server that reverse‑proxies Google Gemini’s web UI protocol and exposes it as an OpenAI‑compatible API (chat, models, streaming, video, image, music, web‑search). Works anonymously or with Google cookies, includes rate‑limiting, proxy & cookie pools, SQLite/MySQL persistence, and a lightweight admin UI.

A Python library for portfolio optimization and risk management built on scikit-learn, providing tools for factor model construction and ML-based model selection.