OpenAlice is a locally‑run AI orchestrator for individual traders. It lets you plug in LLM agents (Claude, Codex, Pi, etc.) and connect them to market data, research workspaces (Git‑backed folders), and broker accounts. Agents can fetch data, run quantitative analyses, generate reports, and even draft trade orders, which you review and commit via a Git‑style interface. The product ships as a desktop app (macOS/Windows) with CLI/Docker options, and includes scheduling, issue tracking, and a central Inbox for follow‑up. Trading execution is marked as beta and intended for paper or test accounts first.

gawkbot is an open‑source platform that lets you describe a workflow in plain language and automatically builds a dedicated AI‑driven micro‑app (bot) to run it. The bots run locally (or on a VM you control), integrate with 1200+ services, and require your explicit approval for any external write. It supports multiple LLM providers (Claude Code, Codex, Ollama, etc.), offers a web UI with live build streaming, scheduling, typed data tables, and citation‑backed knowledge pages, all under a Sustainable Use License.

OpenOutreach is a self‑hosted CLI that uses an LLM and a licensed lead‑data service to discover, qualify, and email B2B prospects based on a simple product description. It bundles two child tools—OpenOutFind (search & qualification) and OpenOutSend (email enrichment & sending)—into a one‑command workflow, requires only an LLM key, a BetterContact API key, and SMTP credentials, and outputs ready‑to‑import CSV files with a human‑readable reason for each lead.

Z3r0 is an AI‑native red‑team workbench that combines a React UI, FastAPI control plane, multi‑agent runtime with LightRAG retrieval, and Docker‑sandboxed execution. It records every action as immutable evidence and provides replayable timelines, making authorized penetration testing auditable and collaborative.

CLI Printing Press is a Go‑based generator that, using Claude Code (or Codex), turns any API or website into a token‑efficient Go CLI plus a matching MCP server. It automatically researches docs, scrapes traffic, builds a local SQLite data layer, adds compound analytics commands, verifies the output, and can publish the result to a public library. Install via a single script, then invoke `/printing-press <api>` inside an AI‑assistant session to get a ready‑to‑use CLI (e.g., ESPN, flight‑goat, linear‑pp-cli) that works offline, outputs compact JSON, and is optimized for AI agents.

Agent of Empires (AoE) is a session manager that runs AI coding agents in persistent tmux sessions on Linux/macOS. It adds a TUI, web dashboard, CLI, and HTTP API, plus Git worktree, Docker/Podman sandboxing, diff viewing, and status detection, letting you monitor and control multiple agents safely and remotely.

Reverify is a Python toolkit that lets LLMs propose facts about binaries (or source code) and then deterministically checks those claims against the actual file. It offers pure‑Python parsers/disassemblers plus optional native back‑ends (Capstone, Unicorn, LIEF, Z3, angr). Verified facts are stored in a ledger, enabling AI agents to roll over contexts without losing trustworthy state. The tool is packaged on PyPI, includes a CLI and an MCP server for agent integration, and has been benchmarked to catch 100 % of hallucinated claims on real Windows binaries.

A curated index of 53 Claude‑compatible *Agent Skills* (design, data, video, writing, persona‑based reasoning, etc.) with install instructions, an update checker, and a machine‑readable catalog. Enables AI agents to perform concrete productivity tasks without custom coding.

ThunderKittens is a header‑only CUDA library that provides a tiny, tile‑oriented DSL for writing fast deep‑learning kernels (e.g., GEMM, FlashAttention). It abstracts tensor‑core calls, async memory moves, and shared‑memory banking while keeping the code compact and type‑safe. Designed for Hopper/Blackwell (and Vera Rubin) GPUs, it lets researchers and engineers produce near‑peak‑performance kernels that can be called from C++ or wrapped for PyTorch.

A local application that converts video audio into multilingual subtitles using whisper.cpp for transcription and Hy-MT2 or cloud APIs for translation.

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.

Magic Context is a CortexKit plugin that gives coding‑assistant agents a self‑managing long‑term memory. It captures useful facts from chat history, consolidates them in the background, and automatically recalls relevant memories across sessions, eliminating the disruptive compaction pauses typical of AI‑coding tools.

A Python MCP server that logs into Garmin Connect and exposes 110+ health, activity, and training tools (read and limited write) for LLMs like Claude. Install via a Claude Desktop `.dxt` extension or run locally; authenticate once with `garmin-mcp-auth` and then LLMs can query steps, sleep, workouts, download FIT files, create and schedule workouts, etc.

svg‑diagram is a zero‑dependency Node skill that defines a strict visual style for SVG diagrams generated by AI coding agents and provides a command‑line linter (`svg‑lint`) to verify that a finished SVG follows the style.

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

CLI Agent Orchestrator (CAO) is a Python tool that runs a local server, launches AI coding CLIs (like Claude Code, Copilot, etc.) in isolated tmux sessions, and provides a unified CLI/Web UI for supervising multiple agents in parallel or sequential workflows. It supports skills, persistent memory, plugins, and even Kubernetes deployment, letting developers orchestrate a team of AI coding assistants.

Skill Seekers is a Python CLI (and optional MCP server) that scrapes 18 kinds of docs/code (websites, GitHub, PDFs, videos, chat exports, etc.), lets an LLM enhance the material into a comprehensive `SKILL.md`, and then packages the result for 22 AI/RAG targets (Claude, Gemini, OpenAI, LangChain, LlamaIndex, Pinecone, etc.). It also installs the packaged skill directly into 19 AI coding assistants (Cursor, Windsurf, Cline, etc.).

Semble is a CPU‑only Python library/CLI/MCP server that indexes a codebase in ~0.5 s and answers natural‑language code queries in ~1 ms, delivering transformer‑level relevance while using ~99 % fewer tokens than a grep‑plus‑read baseline. It combines static code embeddings (Model2Vec) with BM25, fuses results, and applies code‑aware reranking. Integrates with Claude Code, Codex, Cursor, OpenCode, etc., and requires no API keys or GPUs.

Muesli is a native macOS app that provides on‑device speech‑to‑text dictation and meeting transcription. It uses Apple‑silicon‑optimized models (Parakeet, Whisper, Nemotron, etc.) for low‑latency transcription, with optional hosted ASR via OpenAI/OpenRouter. Features include hold‑to‑talk dictation, voice‑driven editing (Quill), VAD‑driven chunking, speaker diarization, AI‑generated meeting notes, iCloud text sync, Siri/Shortcuts integration, and a JSON‑first CLI for agent automation.

A real-time audio translation tool for Windows 11 that integrates Windows Live Captions with LLM and traditional translation APIs to provide translated subtitles via a transparent overlay.

An ultra-low-latency, self-hosted speech-to-text pipeline that enables real-time transcription and translation using advanced simultaneous speech research and multiple model backends.

ReachAI is an open‑source platform that adds AI assistants to existing Java enterprise systems (OA, ERP, CRM, etc.). It scans the app, registers annotated business methods as reusable capabilities, lets LLMs generate the integration code (AI‑Coding), and runs user requests through controlled, versioned workflows with full audit trails. The stack includes Spring Boot, AgentScope, LangGraph4j, Vue front‑end, MySQL/Redis/Milvus, and is MIT‑licensed.

A multi-instrument music transcription model that converts audio recordings into MIDI and professional sheet music using a transformer decoder architecture.

WeSmartFlow is an open‑source, agent‑driven adaptive learning platform (Python 3.10+, Vue 3) that uses LLM agents to understand goals, generate personalized learning paths, create interactive visualisations, quizzes, and HTML knowledge cards, and store progress in a personal knowledge graph. It supports full‑course generation, plug‑in “learning worlds”, and optional WeChat integration via the WeClaw channel. The repo includes a reusable `agent_core` library, FastAPI backend, Vue frontend, and tooling for building and deploying new educational experiences.

A multilingual, tiny speech generation model with 0.1B parameters designed for real-time, CPU-based inference and high-fidelity voice cloning.

An AI-powered multimedia tool that integrates speech recognition, translation, and voice cloning for automated dubbing and multilingual content creation.

A CPU/GPU converter that turns non-DRM eBooks into audiobooks using advanced AI text-to-speech engines with support for voice cloning and 1158 languages.

OpenBiliClaw is a locally‑run, open‑source AI agent that builds a psychological profile of you from your activity on many content platforms (Bilibili, YouTube, Reddit, etc.). Using that profile, it actively searches those sites for items it predicts you’ll like, explains its choices, and keeps all data on your machine. It consists of a browser extension, a Go/Python backend (with optional offline embedding model), a web UI for desktop and mobile, a Flutter native client, and optional DeepSeek Harness integration. Installation is a four‑step process: add the extension, install the backend (installer or AI‑assisted script), log into a source platform, and open the local web UI. The project emphasizes privacy, cross‑platform recommendation, and continuous self‑learning.

js‑reverse‑mcp is a Node.js MCP server that lets LLM agents debug and reverse‑engineer JavaScript in Chrome. It wraps DevTools data (scripts, breakpoints, network, WebSocket, storage) into 24 concise tools designed for AI‑driven reasoning, and includes optional stealth layers (Patchright fork and CloakBrowser) for anti‑bot sites.

Terax is an open‑source, lightweight desktop IDE that combines a GPU‑accelerated terminal, CodeMirror editor, Git UI and a web preview, plus an integrated AI side‑panel that works with cloud LLMs or local models. It offers AI‑driven code suggestions, edit diffs, and agentic workflows, all within a ~7 MB Tauri‑based app for macOS, Linux and Windows.

Snyk Agent Scan is a CLI tool that discovers AI‑agent components (MCP servers, skills, and various agents) on a machine and checks them for security risks like prompt‑injection, secret leakage, and malicious code. It works by auto‑discovering config files, optionally launching MCP servers (with user consent), sending sanitized data to Snyk’s analysis service, and reporting scored risk indicators. Install via `uvx` or a signed binary, set a Snyk API token, and run the scan (preferably in a sandbox).

cue is an open‑source Electron desktop app that floats a transparent panel over your screen, captures your screen, microphone, and (on Windows) meeting audio, and sends those streams to a language model of your choice for real‑time assistance in meetings and coding. It works on macOS 15.4+ and Windows 10 2004+, supports OpenAI, Anthropic, Gemini, Azure, Ollama, or any OpenAI‑compatible endpoint, and can use local Whisper‑cpp for on‑device transcription. The app tries to stay hidden from screen‑share tools via OS‑level exclusion flags, but this is best‑effort. Installation is via pre‑built binaries or `npm` from source; first‑run guides you through OS permissions, API‑key setup, and optional Zoom configuration. No telemetry is collected; keys and optional résumé text are stored locally.

go‑stock is a Windows/macOS desktop app that combines real‑time stock data with LLM‑driven analysis, offering chart tools, AI agents, back‑testing, alerts, and a skill‑sharing system. It works with many LLM providers (OpenAI, local models, Chinese cloud models) and is GPL‑v3 licensed.

Moshi is a full-duplex speech-text foundation model designed for real-time, low-latency spoken dialogue, utilizing a specialized streaming audio codec called Mimi.

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.

LeetCUDA is a GPL‑v3 repository that provides a 400‑page learning book and over 200 CUDA kernels (including high‑performance HGEMM and Flash‑Attention implementations) wrapped for PyTorch, targeting AI workloads on NVIDIA Tensor‑Core GPUs.

A tiny npm proxy that forwards legacy Magic‑MCP calls to the new 21st MCP server, preserving old configs while giving access to the updated AI component generation tools.

herdr‑reviewr is a terminal UI plugin for the *herdr* AI‑assistant that shows live diffs of an agent’s code changes, lets you comment on specific lines, and send those comments back to the agent—all within a split pane or as a standalone app. It supports multiple diff scopes, fuzzy file search, PR read‑only view, theming, and full key‑binding customization.

Sessions Viewer is a cross‑platform desktop app (Tauri + Vue 3) that lets you open, search, replay, and continue local transcript files from AI coding assistants like Claude Code, Codex, Grok Build, Kimi Code, Pi, Antigravity CLI and opencode. It provides faithful replay of tool calls, global search, split‑pane layouts, token‑cost analytics, and one‑click resume of sessions in embedded or external terminals, all while keeping the original JSONL files read‑only.

Clawd on Desk is a cross‑platform desktop pet that visualises the activity of AI coding agents (Claude, Copilot, Gemini, Cursor, etc.) in real time, showing animations for thinking, typing, tool use, permission requests, and task completion. It supports dozens of agents via hook files, offers clickable permission bubbles (with optional Telegram/Feishu/Slack forwarding), a session dashboard, a read‑only mobile PWA, and customizable themes.

A voice input addon for Fcitx5 that enables local or cloud-based speech-to-text with LLM-powered post-processing and text modification.

JimuReport is a free, open‑source, low‑code reporting suite that combines traditional tabular reports, dashboard/big‑screen designers, and a conversational AI BI (ChatBI). With a built‑in Claude‑Code skill, users can generate or edit visualisations from natural‑language prompts or screenshots. It integrates via Spring Boot starters, supports 30+ data sources (including Chinese government‑approved databases), and can be run locally, via Docker, or as a cloud demo.

A PyTorch-based toolkit for developing and deploying professional speech AI models, including ASR, TTS, and Speech LLMs.

M365 Copilot2API is a Go‑based self‑hosted gateway that translates Microsoft 365 Copilot’s private WebSocket protocol into standard OpenAI‑ and Anthropic‑compatible REST endpoints. It provides a familiar `/v1/chat/completions` (and related) API, supports streaming, function calling, multimodal inputs, and includes a web UI for account/OAuth management, API‑key handling, proxy pools, usage stats, and automatic session cleanup. The tool lets existing OpenAI/Anthropic clients (Claude Code, Cursor, etc.) call M365 Copilot without rewriting code, but it is unofficial and intended for personal research only.

A macOS application for real-time audio transcription using Whisper and Parakeet engines, featuring flexible keyboard and mouse triggers for quick voice-to-text conversion.

Better Harness is an open‑source tool that analyses the whole workflow of AI coding agents (Claude Code, Codex, Qoder, Cursor, GitHub Copilot, etc.). It gathers project and session evidence, evaluates five dimensions of the *Agent Work Loop* (task understanding, controlled execution, change validation, reliable delivery, learning capture), and produces a visual HTML report with prioritized, evidence‑backed findings and repair suggestions. Missing evidence is shown explicitly, helping teams improve repeatability, validation, and learning while still benefiting from AI‑generated code.

A local Excalidraw canvas that AI agents can control via CLI, MCP, or REST, enabling programmatic creation, inspection, and version‑controlled export of diagrams.

Mike is an open‑source, self‑hosted legal AI platform that lets users chat with, review, and research legal documents using LLMs (OpenAI, Anthropic, Gemini, or local Ollama). It bundles a Next.js frontend, Express backend, Supabase auth/Postgres, and R2‑compatible storage, and offers connectors for Slack and Microsoft Word.