I package open-source AI models for speech, translation, OCR and talking-head video into all-in-one Docker images with a Web UI, REST API and MCP server, so they're easy to run locally. I also build agent tooling: MCP servers, AI agent skills and a Bash loop that drives a coding agent through PRD user stories. I've built ZK identity protocols on HashKey Chain and send fixes and security reports upstream.
- 🛡️ Credited Reporter of security advisory GHSA-f8v6-r4j4-f7r7 (high) in lexmount/moli — moli serve: web pages could take over the CDP endpoint (no Origin / Host validation) and read local files
- 🔧 4 merged pull requests to 2 other open-source projects:
- lexmount/moli#2293 fix(renderer): never reuse backendNodeIds across Documents
- lexmount/moli#2292 fix(serve): honor --document-start-script in protocol-server pages
- lexmount/moli#2295 fix(system-info): accept the frontend's browser session (SystemInfo always rejected over real connections)
- wiltodelta/remove-ai-watermarks#11 fix(ctrlregen): correct module import paths
cosyvoice-docker — ★89 · Python · Speech & audio AI
All-in-one Docker image serving the Fun-CosyVoice3-0.5B TTS model with a Web UI, OpenAI-compatible REST API and voice cloning.
- OpenAI-compatible /v1/audio/speech endpoint with real PCM streaming at ~1.2s TTFB
- Custom voice management with Fun-ASR-Nano auto-transcription; embedding cache 53% faster after first use
- Supports Chinese, English, Japanese, Korean plus 18 dialects; deploy with one docker run command
Python Docker CosyVoice Fun-ASR-Nano NVIDIA Container Toolkit
translategemma — ★40 · Python · LLM apps
A local AI translation service built on Google's TranslateGemma, offering a Web UI, REST API and MCP integration for 55 languages.
- All-in-one 82GB Docker image with all 6 models (4B/12B/27B, Q4/Q8) embedded
- Smart chunking for long text and streaming translation output via REST API
- MCP integration that works with Claude Desktop and other AI assistants
Python TranslateGemma Docker CUDA GGUF MCP
supertonic-tts-enhanced — ★23 · Python · Speech & audio AI
An enhanced packaging of the Supertonic on-device TTS system that bundles it in a Docker image with a FastAPI server and a Streamlit web UI.
- One-command all-in-one Docker image with GPU (CUDA 12.6.3) or CPU-only mode
- FastAPI /synthesize endpoint on port 8088 with API docs and a health check
- Streamlit UI on port 8501 with 4 voice styles, speed/quality settings, downloads and history
Python ONNX Runtime FastAPI Streamlit Docker CUDA
kokoro-tts — ★19 · Python · Speech & audio AI
All-in-one Docker image for Kokoro-82M text-to-speech with a web UI, REST API, WebSocket, streaming, batch and MCP.
- Supports 8 languages and offers 54+ male and female voices
- CUDA GPU acceleration with idle-timeout memory release, plus a CPU fallback
- MCP server lets AI agents such as Claude use the TTS service
Python Docker CUDA Kokoro-82M WebSocket MCP
MuMuAINovel — ★18 · Python · LLM apps
An AI-powered novel writing assistant that generates outlines, characters, worldbuilding and chapters using OpenAI, Gemini or Claude models.
- Lets you switch between OpenAI, Gemini, Claude and custom API endpoints
- Tracks foreshadowing, checks consistency, scores chapter quality (S/A/B/C/D) and detects duplicate content
- Deploys with Docker Compose on PostgreSQL; README lists 200+ API endpoints
Python FastAPI React PostgreSQL Docker
- 19,838 commits authored in the past year, on 235 different days, across 225 repositories (merge commits excluded; counted from local git history).
- 198 private or unlisted projects — 8 are introduced below, the rest stay unnamed. By domain: Agent tooling (38), Infrastructure (33), LLM apps (29), Content & media (23), Speech & audio AI (18). Languages: Python (90), TypeScript (22), Shell (21), JavaScript (18), HTML (17).
Selected private projects — the code is not public; described here with the owner's permission.
| 🔒 llm-odyssey Content & media A Chinese LLM explainer video series whose code is written by Claude and rendered frame by frame by a custom engine, with frame = f(t).
TypeScript Bun three.js Canvas2D WebCodecs Python |
🔒 mc-agent Agent tooling An AI society experiment where autonomous LLM residents live in a Minecraft Forge world, building, farming and talking without commands.
JavaScript Node.js mineflayer Python Minecraft Forge uv |
| 🔒 mac-h3c Vision & video AI A C/Metal port of MiniMax-H3 for Apple Silicon that makes video and audio from prompts, with frame and reference conditioning.
C Metal FFmpeg Hugging Face |
🔒 TermHub Agent tooling TermHub is an agent-mode terminal hub where Codex watches Claude Code sessions in the terminal and keeps pushing them toward a goal you set.
TypeScript Electron Node.js tmux WebSocket pnpm |
| 🔒 touchstone Agent tooling A Claude Code skill set that teaches how to verify code, not write it, routing to existing skills and filling gaps like mutation checks.
Python Claude Code |
🔒 gba-metro-map Data A code-drawn Greater Bay Area rail map joining MTR, nine-city metros, intercity, HSR and border crossings for route planning.
Python uv OpenStreetMap SVG Vitest Capacitor |
| 🔒 moli-control Browser automation A browser CLI for AI agents that drives Moli headless and managed Chrome with one command set, verifying every action.
Python Rust CDP Chrome Moli |
🔒 chrome-control Browser automation A Python CLI that controls Chrome directly via CDP, including extensions, settings, bookmarks and incognito windows.
Python Chrome DevTools Protocol JSON-RPC websocket-client |
Full list
Python Docker CUDA FastAPI MCP PyTorch Gradio WebSocket TypeScript Solidity JavaScript HashKey Chain vLLM Swagger Flask NVIDIA Container Toolkit ONNX Runtime Docker Compose Node.js Whisper Transformers Hardhat GGUF React HuggingFace Nginx ffmpeg HTML
Every public project, each with a one-line summary.
- expert-knowledge-extraction
★4
An AI agent skill that researches a new domain, designs strategic expert interview questions, and turns expert knowledge into method docs. - kirby
★3· Shell
Kirby is a Bash loop that runs an AI coding agent on PRD user stories, using a fresh AI instance each iteration until every story is done. - open-vibe-island
★1· Swift
Fork of Open Island, a native macOS notch app showing AI coding agent sessions, permission approvals and terminal jump-back. - zfile-mcp-server
★1· Python
An MCP server for ZFile that lets AI assistants list files, upload files, and generate shareable direct or short download links.
5 more
- deep-research-ppt-skill
★1
An AI assistant skill that researches any topic and outputs structured Markdown reports plus a PPT with detailed speaker notes. - hk-horizon-2026-ai-Sentinel-Treasury · TypeScript
AI treasury copilot for HashKey Chain that recommends liquidity, yield and payout splits and executes them on-chain after approval. - hk-horizon-2026-defi-Eternal-Life · Solidity
K-Life is AI agent life insurance on HashKey Chain that watches heartbeats, backs up memory to IPFS and resurrects agents that die. - hk-horizon-2026-ai-OmniAgent · TypeScript
Non-custodial yield router where an AI allocator runs sub-agents that pay each other via X402 and verify risk with ZK proofs. - hk-horizon-2026-ai-Pacifica-Pilot · JavaScript
An AI agent that trades Pacifica perpetuals using Gemini and Elfa sentiment, logging each decision on HashKey Chain.
- translategemma
★40· Python
A local AI translation service built on Google's TranslateGemma, offering a Web UI, REST API and MCP integration for 55 languages. - MuMuAINovel
★18· Python
An AI-powered novel writing assistant that generates outlines, characters, worldbuilding and chapters using OpenAI, Gemini or Claude models. - langextract-web
★9· Python
A web UI, REST API and MCP server wrapper for Google's LangExtract that uses LLMs to pull structured information out of text. - hy-mt
★5· Python
All-in-one Docker deployment of Tencent HunyuanMT 1.5 with a Web UI, REST API and MCP server for translation in 38 languages.
4 more
- translategemma-cli
★3· Python
A local translation service on Google's TranslateGemma, usable via Web UI, REST API or MCP, covering 55 languages. - fabric-docker
★2· Go
An all-in-one Docker deployment of the Fabric AI augmentation framework that runs a Web UI, a REST API, and an MCP server in one container. - PaperRadar
★1· Python
A self-hosted AI research platform that finds new arXiv papers, then translates, analyzes and links them and sends alerts. - EasyPaper
★1· Python
Self-hosted web app that translates or simplifies English academic PDFs, highlights key sentences, and builds a knowledge base.
- cosyvoice-docker
★89· Python
All-in-one Docker image serving the Fun-CosyVoice3-0.5B TTS model with a Web UI, OpenAI-compatible REST API and voice cloning. - supertonic-tts-enhanced
★23· Python
An enhanced packaging of the Supertonic on-device TTS system that bundles it in a Docker image with a FastAPI server and a Streamlit web UI. - kokoro-tts
★19· Python
All-in-one Docker image for Kokoro-82M text-to-speech with a web UI, REST API, WebSocket, streaming, batch and MCP. - soulx-podcast-docker
★18· Python
Docker deployment of SoulX-Podcast-1.7B for multi-speaker podcast speech, served via web UI, REST API and MCP server.
28 more
- indextts2-docker
★17· Python
Docker images for IndexTTS2 that run zero-shot text-to-speech with emotion control, available through a REST API and a Gradio WebUI. - qwen3-tts
★16· Python
All-in-one Docker deployment of Qwen3-TTS with a Web UI and REST API for custom-voice, voice-design and voice-clone speech. - Qwen3-ASR
★11· Python
All-in-one Docker deployment of Qwen3-ASR for 52-language speech recognition with a web UI, REST API and MCP server. - personaplex
★11· Python
All-in-one Docker setup for PersonaPlex, a speech-to-speech model for real-time full-duplex voice chat with role and voice control. - soniox-asr-web
★7· JavaScript
A Soniox API web app that transcribes audio/video files and does real-time speech recognition and translation over WebSocket. - glm-asr
★7· Python
A speech recognition service built on GLM-ASR-Nano, offering a Web UI, a REST API with SSE streaming, and segmentation for long audio. - fun-asr-docker
★7· Python
An all-in-one Docker image that serves Alibaba's Fun-ASR-Nano-2512 speech recognition model through a Web UI, a REST API, and WebSocket. - faster-whisper-web
★6· Python
A GPU-accelerated speech transcription service built on faster-whisper, with a web UI, streaming output and an OpenAI-compatible API. - mlx-audio
★5· Python
A text-to-speech and speech-to-text library and web service built on Apple's MLX framework that streams generated audio on Apple Silicon. - Step-Audio-R1.1
★4· Python
All-in-one Docker deployment of the Step-Audio-R1.1 audio language model with vLLM, a Web UI and an OpenAI-compatible API. - Fun-Audio-Chat
★4· Python
All-in-one Docker image for the Fun-Audio-Chat audio language model, with a bilingual web UI, REST API and MCP support. - remove-noise-service
★4· Python
A Dockerized, GPU-accelerated service that removes noise from audio files using the DeepFilterNet model, with a Swagger-documented HTTP API. - whisperlivekit-enhanced
★4· Python
Enhanced WhisperLiveKit: a self-hosted, low-latency speech-to-text server with lazy model loading and automatic GPU memory release. - higgs-audio-web
★3· Python
An all-in-one Docker deployment for Higgs Audio v2 that provides a Web UI, REST API and MCP server for expressive TTS and voice cloning. - chatterbox-tts-docker
★3· Python
All-in-one Docker image for Resemble AI's Chatterbox TTS with a Web UI, REST API, WebSocket and GPU memory management. - kyutai-tts-docker
★3· Python
A Docker deployment of Kyutai TTS that serves text-to-speech through a web UI, a REST API and MCP tools, and manages the GPU for you. - kyutai-tts-gpu-resident
★2· Python
A Dockerized Kyutai TTS service that keeps the model in GPU memory, with 584 voices, voice cloning and streaming via web UI and REST API. - VibeVoice
★2· Python
Packaging of Microsoft's VibeVoice that serves real-time streaming text-to-speech over WebSocket from a Docker image. - ringtone-forge
★1· Python
A Python CLI and agent skill that turns a song into a 30-second ringtone by finding its chorus and applying a genre-adaptive envelope. - edgetts-ws
★1· Python
A lightweight aiohttp HTTP API that wraps Microsoft Edge TTS and returns MP3 audio plus per-word timestamps for word-by-word highlighting. - fish-speech
★1· Python
A multilingual text-to-speech system with reusable speaker voice profiles, auto-transcription and emotion control via REST API and WebUI. - IMS-Toucan
★1· Python
A Docker-packaged build of IMS-Toucan, the University of Stuttgart's multilingual text-to-speech system, with a Web UI and REST API. - orpheus-tts-docker
★1· Python
Docker deployment for Orpheus TTS that serves text-to-speech through a Web UI, a REST API and MCP, and manages GPU loading for you. - Step-Audio2
★1· Python
Docker-ready deployment of the Step-Audio 2 multimodal LLM for audio understanding and speech chat, with smart GPU memory management. - llasa-tts-8b-webui
★1· Python
A text-to-speech system built on Llasa-8B that supports voice cloning and is served through a Gradio web UI, a Flask REST API and MCP. - pyvideotrans
★1
An AI-powered tool that translates and dubs videos. - Step-Audio-EditX-GPU-Managed
★1· Python
Step-Audio-EditX audio editing model with added GPU memory management that loads models on demand and offloads them when idle. - Step-Audio-EditX-Optimized · Python
Optimized deployment of Step-Audio-EditX, a 3B LLM audio editing and zero-shot TTS model, with UI and API sharing one model instance.
- PersonaLive
★12· Python
A live-streaming portrait animation tool that animates a reference portrait from a webcam in real time or from a driving video offline. - sharp
★8· Python
All-in-one Docker for Apple's SHARP model that turns one photo into a 3D Gaussian Splat scene, via Web UI, REST API or MCP. - infinitetalk-deployment
★8· Python
Docker and Gradio deployment for MeiGen-AI's InfiniteTalk and MultiTalk audio-driven video models, with auto model download. - StableAvatar
★5· Python
Docker deployment of StableAvatar that turns one image plus audio into a talking-head video of any length.
10 more
- HunyuanOCR-WebUI
★4· Python
A Gradio web UI for Tencent's 1B-parameter HunyuanOCR model for text spotting, document parsing, info extraction and image translation. - monkeyocr-multimodel
★3· Python
A MonkeyOCR fork that adds Qwen2.5-VL-3B and Qwen3-VL-8B support to document parsing for handwritten ancient Chinese recognition. - seedvr2-docker-allinone
★3· HTML
A Docker all-in-one build of ByteDance SeedVR2 that upscales videos and images through a bilingual web UI with a task queue. - depth-pro-docker
★2· Python
Docker deployment of Apple's Depth Pro for single-image metric depth estimation, with a Web UI, REST API and MCP server. - ditto-talkinghead
★2· Python
A motion-space diffusion system that turns a single portrait image and an audio clip into a realistic, controllable talking head video. - Z-Image-Studio
★1· Python
An all-in-one Z-Image-Turbo toolkit for model downloads, multi-LoRA inference, LoRA training, and web UI management. - Step1X-Edit-Docker
★1· Python
A Docker deployment of the Step1X-Edit image-editing model that offers a UI, an API and an MCP server, and manages GPU memory automatically. - ovis-image-docker
★1· Python
Docker deployment for the Ovis-Image 7B text-to-image model that serves a Web UI, REST API and MCP server from one container. - yolo11x-stream-detection
★1· Python
A Dockerized system that runs Ultralytics YOLO11x detection on live RTMP video streams and serves annotated output over HLS. - dots-ocr
★1· Python
A ~1.7B-parameter multilingual vision-language model that parses document layouts and recognizes their content.
- electrify-web
★3· TypeScript
A CLI tool that turns any website into a desktop app with one command, built as a modern revival of the abandoned Nativefier project. - cursor-session-manager
★1· HTML
A Cursor AI IDE tool that saves, restores, searches and auto-saves Cursor Agent conversation history via a web UI or CLI.
- lean-allinone
★6· Python
All-in-one Docker image for the QuantConnect Lean trading engine with GPU support, a Web UI, a REST API and an MCP server. - kimi-linear-vllm-docker-serve
★2· Python
Docker image that serves Kimi-Linear-48B-A3B (AWQ-4bit) via vLLM with an OpenAI-compatible API and 128K–1M context. - f-stack
★1· C
A copy of Tencent Cloud's F-Stack, a DPDK framework running the FreeBSD 13.0 network stack in user space, with added reports and docs. - gpu-container-monitor
★1· HTML
A real-time web dashboard for monitoring NVIDIA GPUs and managing GPU-enabled Docker containers.
3 more
- hk-horizon-2026-zkid-Veritier · TypeScript
A privacy-preserving tiered identity and access protocol for humans and AI agents on HashKey Chain, using Groth16 ZK proofs. - hk-horizon-2026-zkid-Aria-Protocol · JavaScript
Compliance middleware on HashKey Chain combining ZK-KYC, AI risk scoring, ERC-4626 vaults and HSP settlement for institutional DeFi. - quickwit-benchmark · Python
A Locust-based load testing project that measures query performance of the Quickwit logging system under concurrent users.
- deepwiki-scraper
★4· JavaScript
A tool that scrapes DeepWiki project documentation and merges it into a single Markdown file, via a command line or a web interface. - tradingview-api-python
★1· JavaScript
A Python FastAPI service serving TradingView real-time market data, technical analysis and indicators via REST. - hk-horizon-2026-defi-Critmin-Oracle · Python
An on-chain oracle on HashKey Chain Testnet that publishes AI-computed risk scores for lithium, nickel and cobalt for DeFi use.
- CoolSnakeGame
★1· Python
A cyberpunk-styled neon snake game in Python and Pygame where you can play solo, watch an AI play, or compete against it.
- notebookLM2PPT
★17· Python
A Python tool that converts PDF slides like NotebookLM exports into vector-graphics PowerPoint files via CLI, web UI, API or MCP. - google-flow-downloader
★2· Python
A Python CLI that batch downloads AI-generated images from Google Flow via a browser-exported JSON file or direct API access. - calendar-format
★1· Python
A Python CLI that merges Taoist, Buddhist and Hong Kong holiday data into a landscape A4 calendar PDF, one page per month. - pdf2ppt
★1· Python
A Python tool that converts PDF slides into PowerPoint presentations while keeping vector graphics at the highest resolution.
- polycopy
★6· TypeScript
A TypeScript bot that copies trades from chosen Polymarket traders, with risk controls, a web dashboard, REST API and MCP server. - FanPilot
★2· C
A macOS menu bar app and daemon for Apple Silicon that holds fans at a set percentage and runs them at full speed when the Mac overheats. - looptrade
★1
New loop project - python3
★1
all_python
13 more
- insight
★1 - hk-horizon-2026-zkid-ZKGate · TypeScript
A HashKey Chain protocol that lets users prove KYC, age or accredited status to DeFi apps via ZK proofs without revealing personal data. - hk-horizon-2026-defi-eldra · TypeScript
Eldra is a monorepo for an RWA protocol covering fundraising, fractional ownership, secondary trading and liquidity rewards. - hk-horizon-2026-defi-hash-credit · Solidity
HashCredit is a HashKey Chain protocol that turns SPV-proven Bitcoin mining payouts into credit lines for borrowing stablecoins. - dots.ocr
- soulbridge
SoulBridge is an AI application for practicing languages. - crawler_demo2 · HTML
- crawler_demo
- MWAA_best_practice
- emr-jupyterlab
emr-jupyterlab - guetzli
mytest git - HPT
Real estate block chain - Shamir-Secret-0th-test · Python
A Python command-line PoC that splits secrets like Ethereum, Bitcoin or login keys into shares using Shamir's Secret Sharing.
And 198 more private or unlisted projects.
🥚
- 19,841 commits since 2025-04 (merge commits excluded)
- busiest day: 2026-10-21, with 1,004 commits
- longest streak: 97 days in a row
- 33% of commits land between 10 pm and 5 am; favourite hour: 00:00
- if every commit were a step, that walk would be about 15.1 km
This page is generated from repository data by a private tool. Every repository linked here is re-verified public before publishing; counts and charts are recomputed from the same data, never hand-written.