Every repo here was really executed in our locked-down sandbox. The rank comes from evidence, not stars: did it install cleanly, do its own tests pass, does its example actually run.
Updated automatically as the lab tests — last build July 14, 2026 · 100 ranked · ribbons show what a thing IS and what it runs on (proven vs. declared).
Spec Kit is a toolkit for Spec-Driven Development (SDD) that focuses on creating executable specifications rather than just code.
Installed cleanly on the first try; its own test suite ran — 3,908 tests passed; the demo actually ran and produced real output.
Hivemind is a cloud-backed shared memory and skill-learning system for AI agents.
Its own test suite ran — 5,189 tests passed.
An unofficial Python API and agentic skill that provides programmatic access to Google NotebookLM's features.
Its own test suite ran — 10,615 tests passed.
Click is a composable command-line interface (CLI) toolkit for Python.
Installed cleanly on the first try; its own test suite ran — 1,653 tests passed.
HTTPX is a next-generation HTTP client for Python that provides both sync and async APIs.
Its own test suite ran — 1,413 tests passed.
A persistent memory system for AI coding agents that provides long-term context retention across sessions.
Its own test suite ran — 1,410 tests passed.
AgentScope is a production-grade multi-agent framework designed for building and deploying agentic LLM applications.
Installed cleanly on the first try; its own test suite ran — 1,239 tests passed.
A local proxy and trace viewer that intercepts and inspects API traffic from various AI coding agents (e.g., Claude Code, Cursor, Gemini CLI).
Installed cleanly on the first try; its own test suite ran — 865 tests passed; the demo actually ran and produced real output.
Quoroom is an open-source swarm intelligence engine that enables the creation of self-governing AI collectives.
Its own test suite ran — 1,091 tests passed.
Toolscore is a Python framework for evaluating LLM tool-calling behavior using deterministic, local tests.
Its own test suite ran — 870 tests passed; the demo actually ran and produced real output.
PentestAgent is an AI agent framework designed for automated black-box security testing, including bug bounty and red-team workflows.
Installed cleanly on the first try; its own test suite ran — 608 tests passed; the demo actually ran and produced real output.
CORAL is an infrastructure for multi-agent autonomous self-evolution and autoresearch.
Its own test suite ran — 592 tests passed; the demo actually ran and produced real output.
Helix is a self-healing runtime that wraps function calls (like payments or API requests) to automatically diagnose, fix, and remember errors.
Installed cleanly on the first try; its own test suite ran — 632 tests passed.
Requests is a highly popular and elegant HTTP library for Python that simplifies making HTTP/1.1 requests.
Installed cleanly on the first try; its own test suite ran — 615 tests passed.
NanoClaw is a lightweight AI agent framework that allows users to run Claude agents in isolated Linux containers.
Installed cleanly on the first try; its own test suite ran — 588 tests passed.
Crune is a toolset that analyzes Claude Code session logs to generate a semantic knowledge graph and reusable skill definitions.
Installed cleanly on the first try; its own test suite ran — 520 tests passed.
MARM is a self-hosted memory server for AI agents that provides persistent context across sessions and projects using a Model Context Protocol (MCP) i.
Its own test suite ran — 604 tests passed.
A comprehensive Douyin downloader that supports downloading videos, image-notes, music, and entire user profiles.
Installed cleanly on the first try; its own test suite ran — 359 tests passed.
Agience is an operating system for AI workflows that transforms AI-generated output into durable, governed knowledge artifacts with unique identities .
Installed cleanly on the first try; its own test suite ran — 347 tests passed.
TensorTrade is an open-source reinforcement learning (RL) framework designed for building, training, and evaluating algorithmic trading agents.
Its own test suite ran — 241 tests passed; the demo actually ran and produced real output.
A post-hoc debugger for LLM reasoning traces that parses chain-of-thought transcripts into directed graphs.
Installed cleanly on the first try; its own test suite ran — 113 tests passed; the demo actually ran and produced real output.
Hound is an AI-powered security auditing tool that builds and refines adaptive knowledge graphs to perform deep code reasoning.
Installed cleanly on the first try; its own test suite ran — 92 tests passed; the demo actually ran and produced real output.
Verdict is a declarative framework for building complex 'LLM-as-a-judge' systems by scaling up judge-time compute.
Installed cleanly on the first try; its own test suite ran — 87 tests passed; the demo actually ran and produced real output.
Menelaus is a Python library providing algorithms for detecting concept and data drift in machine learning models.
Its own test suite ran — 164 tests passed; the demo actually ran and produced real output.
A Streamlit application for generating identity-preserving portraits that allows users to compare multiple fine-tuning strategies and backbones (e.g.,.
Installed cleanly on the first try; its own test suite ran — 60 tests passed; the demo actually ran and produced real output.
LoopCraft is a runtime harness and plugin for the Hermes AI agent framework that implements 'loop engineering.' It wraps tool execution with a structu.
Installed cleanly on the first try; the demo actually ran and produced real output.
A high-performance navigation stack for multi-robot systems that combines Conflict-Based Search (CBS) for global path planning with Optimal Reciprocal.
Installed cleanly on the first try; the demo actually ran and produced real output.
Elume is a cognitive substrate for AI agents that provides a deterministic, replay-safe memory and mental-modeling architecture.
Installed cleanly on the first try; the demo actually ran and produced real output.
A Model Context Protocol (MCP) server that enables Large Language Models (LLMs) to interact with Google Analytics APIs.
Installed cleanly on the first try; the demo actually ran and produced real output.
Tenacity is a general-purpose retrying library for Python that allows developers to retry functions that fail due to transient errors.
Installed cleanly on the first try; the demo actually ran and produced real output.
HiddenBench is a benchmark designed to evaluate the collective reasoning capabilities of multi-agent LLM systems using the 'Hidden Profile' paradigm.
Installed cleanly on the first try; the demo actually ran and produced real output.
SkillOpt is a text-space optimization framework that treats natural-language agent skills as trainable states.
Installed cleanly on the first try; the demo actually ran and produced real output.
Reasoning Gym is a Python library providing procedural dataset generators and algorithmically verifiable reasoning environments for training models vi.
Installed cleanly on the first try; the demo actually ran and produced real output.
A Model Context Protocol (MCP) server that provides a bridge between AI assistants (like Claude) and the Google Sheets API.
Installed cleanly on the first try; the demo actually ran and produced real output.
A toolkit for generating high-quality synthetic datasets to fine-tune Large Language Models (LLMs).
Its own test suite ran — 96 tests passed; the demo actually ran and produced real output.
An MCP (Model Context Protocol) server that provides AI models with tools to interact with Google Docs, Sheets, Drive, Gmail, and Calendar.
Its own test suite ran — 288 tests passed.
An automated short video generation platform that converts text prompts or keywords into high-definition videos.
Installed cleanly on the first try; its own test suite ran — 176 tests passed.
A Python framework for state-space modeling, simulation, estimation, and control of physical systems.
Installed cleanly on the first try; its own test suite ran — 124 tests passed.
IntellAgent is a multi-agent framework designed to stress-test and optimize conversational AI agents by simulating thousands of realistic user interac.
Installed cleanly on the first try; its own test suite ran — 43 tests passed.
A Python framework that implements multiple iterative and adaptive reasoning patterns (AoT, L2T, GoT, FaR) for Large Language Models.
Its own test suite ran — 140 tests passed.
A Model Context Protocol (MCP) server that provides local, graph-based multi-step reasoning and memory management.
Installed cleanly on the first try; its own test suite ran — 20 tests passed.
A microservices-based Retrieval-Augmented Generation (RAG) system built in Go for processing and querying complex documents.
Installed cleanly on the first try.
Paca is an AI-native, self-hosted project management platform designed for Scrum teams where AI agents act as first-class teammates.
Installed cleanly on the first try.
Sentinel is an edge computing agent for Kubernetes that provides predictive failure detection and partition-resilient orchestration.
Installed cleanly on the first try.
Dagrs is a high-performance Rust framework for asynchronous task orchestration using Flow-Based Programming (FBP).
Installed cleanly on the first try.
A multi-agent AI orchestration platform that simulates collective reasoning by organizing a 'council' of specialized AI personas.
Installed cleanly on the first try.
mtix is a hierarchical task management system designed for multi-agent LLM development.
Installed cleanly on the first try.
OpenLIT is an open-source observability and management platform for AI engineering.
Installed cleanly on the first try.
Toil is a file-defined workflow orchestrator written in Go that manages complex, role-driven processes.
Installed cleanly on the first try.
Traceseal Observe provides a cryptographic framework for generating signed execution receipts for AI model calls, tool invocations, and data flows.
Its own test suite ran — 85 tests passed.
Promptify is a task-oriented NLP engine that provides a high-level API for extracting structured data from LLMs using Pydantic schemas.
Its own test suite ran — 78 tests passed.
A curated repository of resources, tools, and best practices for using Anthropic's Claude Code agent.
Its own test suite ran — 41 tests passed.
A community-curated directory of free public APIs across various domains.
Its own test suite ran — 31 tests passed; the demo actually ran and produced real output.
Dagra is a C++-based task orchestrator that executes tasks as a Directed Acyclic Graph (DAG).
The build and test suites completed successfully with 100% of tests passing.
Docs2KG is a tool for constructing unified Knowledge Graphs (KGs) from heterogeneous documents (PDF, DOCX, HTML, EPUB) using a human-LLM collaborative.
Installed cleanly on the first try; the demo actually ran and produced real output.
SolidiFI is an automated framework designed to evaluate smart contract static analysis tools by systematically injecting known bug types into contract.
Installed cleanly on the first try; the demo actually ran and produced real output.
STFlow is a generative framework that uses flow matching to predict spatial transcriptomics (gene expression) from histology images.
Installed cleanly on the first try; the demo actually ran and produced real output.
SE-Agent is a self-evolution framework for LLM-based code agents that optimizes multi-step reasoning trajectories.
Installed cleanly on the first try; the demo actually ran and produced real output.
AGiXT is an AI Agent Automation Platform designed to orchestrate complex tasks across multiple AI providers.
Installed cleanly on the first try; the demo actually ran and produced real output.
AgentClaw is a declarative agent workflow framework that allows users to build and deploy AI agents using a convention-over-configuration approach.
Installed cleanly on the first try; the demo actually ran and produced real output.
Bandit is a security-focused static analysis tool designed to find common security vulnerabilities in Python code.
Installed cleanly on the first try; the demo actually ran and produced real output.
A comprehensive collection of political propaganda and informational resources regarding the Chinese government.
Installed cleanly on the first try; the demo actually ran and produced real output.
VisuoThink is a framework that enhances Large Vision-Language Model (LVLM) reasoning by integrating visuospatial and linguistic domains.
Installed cleanly on the first try; the demo actually ran and produced real output.
Freqtrade is an open-source cryptocurrency trading bot written in Python that supports multiple exchanges and provides a web UI and Telegram integrati.
Installed cleanly on the first try; the demo actually ran and produced real output.
oMLX is an LLM inference server optimized for Apple Silicon that features continuous batching and a tiered KV cache (hot in-memory and cold SSD).
Installed cleanly on the first try; the demo actually ran and produced real output.
Agent S is an open-source framework for creating autonomous GUI agents that interact with computers like humans.
Installed cleanly on the first try; the demo actually ran and produced real output.
A comprehensive Python library for anomaly detection that supports multiple data types including tabular, time series, graph, text, image, and audio.
Installed cleanly on the first try; the demo actually ran and produced real output.
Pixelle-Video is an AI-powered video generation engine that automates the entire production pipeline from a single text prompt.
The demo actually ran and produced real output.
Mythril is a security analysis tool for EVM bytecode that identifies vulnerabilities in smart contracts.
The demo actually ran and produced real output.
MemoGraph is an intelligent memory management system that transforms markdown notes into a queryable knowledge graph for Large Language Models (LLMs).
The project has a complete structure with tests, documentation, and a published package, and the `basic_usage.py` demo successfully indexed and retrieved context.
A framework for creating verifiable execution traces for multi-step AI agent pipelines using ZK-style concepts.
The demo actually ran and produced real output.
PraisonAI is an AI agent framework designed to build autonomous, self-improving agents capable of research, planning, and code execution.
The demo actually ran and produced real output.
OpenAdapt is an open-source framework that bridges Large Multimodal Models (LMMs) with desktop and web GUIs.
The demo actually ran and produced real output.
Prefect is a workflow orchestration framework for building and managing resilient data pipelines in Python.
The demo actually ran and produced real output.
A library for creating environments to train and evaluate Large Language Models (LLMs) using Reinforcement Learning.
The demo actually ran and produced real output.
Skyvern is an AI-powered browser automation platform that uses Large Language Models (LLMs) and computer vision to navigate websites.
The demo actually ran and produced real output.
A data generation tool for Supervised Fine-Tuning (SFT) that uses OpenAI's Chat Completion API to generate synthetic datasets.
Installed cleanly on the first try; the demo actually ran and produced real output.
G-Force is an orchestration engine that automates the deployment of ephemeral Google Cloud Platform infrastructure for machine learning tasks.
The demo actually ran and produced real output.
FastAPI is a high-performance web framework for building APIs with Python based on standard type hints.
The demo actually ran and produced real output.
Harbor is a framework designed for evaluating and optimizing AI agents and language models.
Installed cleanly on the first try; the demo actually ran and produced real output.
Textual Localization is a text-to-image diffusion framework designed to handle multi-concept input images for subject-driven generation.
The demo actually ran and produced real output; the library imports without errors.
Khoj is an AI-powered 'second brain' that allows users to interact with their personal documents and the web using LLMs.
The demo actually ran and produced real output.
Kuwala is a no-code data platform that provides a unified workspace for building analytics workflows.
The demo actually ran and produced real output.
Letta is a platform for building stateful AI agents equipped with advanced long-term memory and the ability to learn and self-improve over time.
The demo actually ran and produced real output.
Katalyst is a modular, node-based terminal coding agent that uses a Directed Acyclic Graph (DAG) structure for task planning.
The demo actually ran and produced real output.
Cairn is a general-purpose state-space search engine that uses a Blackboard Architecture and a fact-intent graph to solve problems with a clear start .
The demo actually ran and produced real output.
Flask is a lightweight Python micro-framework designed for building web applications.
The demo actually ran and produced real output.
Qodo-Cover is an AI-powered tool designed to automatically generate and enhance unit tests to improve code coverage.
The demo actually ran and produced real output.
BatonTrace is an MCP (Model Context Protocol) server designed for high-integrity AI agent orchestration.
The demo actually ran and produced real output.
Typer is a Python library for building command-line interfaces (CLIs) based on type hints.
The demo actually ran and produced real output.
A full-stack web application for multi-speaker voice generation and cloning based on Microsoft's VibeVoice model.
The demo actually ran and produced real output.
EVA is an AI-assisted penetration testing agent that automates and guides offensive security workflows.
Installed cleanly on the first try; the demo actually ran and produced real output.
A Python library for implementing Bayesian Knowledge Tracing (BKT) and its variants to estimate student mastery from problem-solving sequences.
Installed cleanly on the first try.
A multi-disciplinary initiative and dataset collection that audits and documents the web and machine sources, licenses, and metadata of over 1,800 fin.
Installed cleanly on the first try.
OpenDriveVLA is an end-to-end autonomous driving framework that leverages a Large Vision Language Action Model (LVLM).
Installed cleanly on the first try.
SGI-Bench is a scientific evaluation framework designed to measure 'Scientific General Intelligence' in LLMs by simulating the full inquiry cycle: Del.
Installed cleanly on the first try.
A Graph Retrieval-Augmented Generation (Graph RAG) framework that combines graph-based knowledge retrieval with Generative AI.
Installed cleanly on the first try; the demo actually ran and produced real output.
CodeSpace is a distributed remote code execution (RCE) platform designed for real-time educational environments.
Installed cleanly on the first try.
LeanRAG is a Retrieval-Augmented Generation (RAG) framework that uses knowledge graphs to improve response quality and efficiency.
Installed cleanly on the first try.
EnvFactory is a framework for scaling tool-use agents by synthesizing executable environments and robust Reinforcement Learning (RL).
Installed cleanly on the first try.