Repository index
1016 AI, ML and Web3 projects · page 7 of 11
- MOSS-AUDIO: MODEL FOR AUDIO ANALYSIS
MOSS-Audio — an open-source Python model for generating captions, answering questions about sound, and reasoning about speech, environment and music. With 655 stars on GitHub supports audio captioning, audio‑VQA and multimodal understanding.
- Any topic in video
Creates video from text with TTS and subtitles. Remotion generates video in Chinese and English.
- FRAME VIDEO EDITOR FOR AI
cartcut — video editor for AI agents, built on Electron and TypeScript. Uses ffmpeg for processing, Lit for UI, and supports a macOS app. Demonstrates that open source can outperform commercial tools.
- BLITZ STRIKE PENTEST TOOLKIT
⚡ Blitz Strike — universal pentest toolkit for MCP. Methodology: reconnaissance, attack‑surface mapping, source‑to‑sink analysis, live validation. 57 escalation chains, 130‑tool catalog, intelligence data layer; one server — for all agents.
- MOMO CODE: AI AGENT FOR CODE
MOMO CODE — AI agent for programming in TypeScript with CLI. It evolves together with you, adapting to your coding style and suggesting improvements in real time.
- MIXING RECURSIONS FOR LLM
The Mixture-of-Recursions project offers dynamic recursion depth selection per token in a transformer, enabling adaptive compute allocation. This reduces KV‑cache costs and speeds up inference without losing quality.
- SALES DESIGN FOR SAAS
Python library with UI‑templates based on behavioral science and CRO to boost SaaS conversion. Includes A/B‑scripts and hints from Claude and Codex for AI‑design.
- HYPERFRAMES: AI VIDEO EDITOR
Edit videos, Reels and YouTube Shorts via Codex or Claude Code. 14 skills, transcript‑driven cuts, storytelling, 406 motion‑graphics cards and HyperFrames templates.
- AGENT MEMORY: MARKDOWN IN CLOUD
Long‑term memory for AI agents: Markdown — source of truth, local ranked retrieval and a separate sleep layer. Claude Code and Codex share one store, no API key needed.
- KANI: MICROFRAMEWORK FOR LLM TOOLS
Kani — a Python microframework for building tool‑calling agents with support for GPT‑4, Llama and other LLMs. It lets you quickly connect arbitrary functions and tools via a simple API. Developed within NLP‑OSS and presented at EMNLP 2023.
- DIGITAL AGENT SYRENE
Cyrene-Agent — an open‑source desktop AI agent with Cyrene’s persona, built on the Cyrene_Harness framework. It blends lifelike companion chat with practical features: coding help, learning assistance, everyday tasks, music and weather.
- ORCALITE LLM ROUTER SELF-HOSTED
Self-hosted LLM router with managed safety net and OpenAI‑compatible API. Supports BYOK, a single workspace, and streaming generation. For advanced routing use the hosted OrcaRouter.
- TOURMIND: AI-AGENT FOR HOTEL BOOKING
AI agent for end-to-end hotel booking. Compares live rates, verifies availability, books stays and manages reservations via TourMind API.
- BKIT VIBECODE KIT: AI‑NATIVE DEVELOPMENT
bkit combines PDCA and Claude Code for AI‑native development in JavaScript. Kit provides templates, prompts and CLI utilities that accelerate the PDCA cycle and support multilingual AI‑agent generation.
- HARNESSROUTER UNIFIED API FOR AGENTS
HarnessRouter Community Edition — implementation under Apache‑2.0 of a unified API for agents. Runs Codex, Claude Code, Hermes, PI, DSH through a single API with sessions, streaming, files, cancellation and error handling, implementing the UHP standard.
- memoria: SECURE MEMORY FOR AI AGENTS
memoria is a Rust-based solution for secure memory management in AI agents. It ensures data integrity, reduces hallucinations, and maintains long-term context in conversations.
- WAYLAND AI AGENT FOR DESKTOP
Wayland — AI agent on TypeScript, which through Electron perceives the environment, reasons with Claude Code and performs actions in real time. The project uses ACP to orchestrate tasks, evolves thanks to feedback and RAG memory.
- AGENT FOR MACOS 26 AI
Agent — AI assistant for macOS 26, controlling the desktop via local and cloud LLMs. Enables task automation, writing scripts and code, interacting with applications. Supports over 18 model providers.
- MOSS-VL: 11B MODEL FOR VIDEO
The open-weight MOSS-VL series of 11-billion-parameter models is designed for understanding long-form and streaming video in real time. It supports multimodal input and efficient stream processing.
- L HARNESS – SELF-HOSTED CRM FOR LINE
L Harness — self-hosted CRM for LINE Official Account with step-by-step messaging, bulk broadcasts, forms and reservations. Includes MCP and TypeScript SDK for customization. Developed by Shudesu (野田修一).
- RELAX: ASYNCHRONOUS RL FOR MULTIMODAL POST‑TRAINING
Relax — an asynchronous reinforcement learning engine for scalable post‑training agents. It supports efficient distributed training, multi-agent scenarios, and working with various modalities via GRPO.
- CHAT ON STEROIDS: POWERFUL LLM
The project provides local Model Context Protocol capabilities for ChatGPT, fully integrated with the Chrome extension. It supports the Goal, Compact & Resume modes and enables steadfast multi‑agent workflows.
- CONTROL CODE FROM YOUR POCKET
9remote — terminal in your pocket: control Claude Code, Codex, Gemini CLI and your Mac/Linux/Windows from any phone or browser. Code in vibe mode anywhere, no extra software needed.
- UTOPIA: SELF-HOSTED KNOWLEDGE BASE
Utopia — self-hosted platform for RAG over documents on a knowledge graph that stores timestamps for each fact. Implemented in Rust + PostgreSQL, a single binary.
- FABRIC CREATIVE 3D CODING
Fabric is a Node‑based environment for creative coding, 3D graphics and image processing, inspired by Quartz Composer. In Swift it uses Metal for fast rendering and supports LLM prompts.
- HERMES OFFICE: AI SUITE FOR DOCUMENTS
HermesOffice is an AI‑native office suite (Docs, Sheets, Slides, PDF) with the Hermes Agent as its brain. It preserves document bytes unchanged, works fully locally, and supports open standards.
- DOCUMENTATION INTO TESTS AUTOMATIC
Truecourse automatically turns documentation into executable tests. When a test fails, it pinpoints the exact docs section where the product differs from the description, helping quickly spot the mismatches between code and the documentation.
- REVERIFY: VERIFIED REVERSE ENGINEERING
reverify is an AI reverse engineering tool that uses binary analysis tools and verifies model results against the binary, eliminating hallucinations. Written in Python, it integrates with Codex and Claude.
- LOCKKNIFE ULTIMATE TOOL
LockKnife combines TUI and headless CLI for deep Android research. Powered by Python orchestration and Rust core, it works with AI agents, credential cracking, APK analysis and runtime inspection.
- ANDROID TABLET INTO SMART HOME
Ava-Pro turns any Android 5+ tablet into a voice-first Home Assistant kiosk. Native C++ lets 10-year-old devices listen, speak, and control the house with AirPlay-grade synced audio in every room.
- MCP-SECURITY: AI-AGENTS FOR SOC
MCP-Security is an open Python framework for integrating AI agents into SOC/SIEM/SOAR. Automates threat analysis and incident response using LLM.
- TRINITY: PLATFORM FOR AI AGENTS
Trinity is a self‑hosted platform for running AI agents Claude Code, Codex and Gemini. It provides orchestration, versioning and access‑policy management. Licensed under Apache 2.0 and implemented in Python.
- VIDEO RECAP WITH CLAUDE CODE
The project provides Claude Code skills for creating Chinese narration recaps from video files. It exports the result as an editable JianYing/CapCut draft. It runs locally, with no external dependencies.
- AWESOME LIST FOR CODEX CLI
This curated list of over 150 tools, skills, sub‑agents and plugins for OpenAI Codex CLI. Lets you quickly find and add extensions for code generation, refactoring and testing. All links are verified and up‑to‑date.
- RSSH: SMART SSH CLIENT FOR ALL.
RSSH — cross-platform SSH client in Rust with an AI agent for command autocompletion and log analysis. Supports still mobile devices and desktops, providing a unified experience. 495 stars on GitHub.
- CLAUDE CODE WIDGET TRACKING
The Windows tray widget shows current usage limits for Claude Code, Codex, Cursor and other AI assistants. It displays time until limit reset, is free and open source in Rust.
- OPEN-PROMPT-INJECTION
Open-Prompt-Injection — benchmark for evaluating attack and defense against prompt injections in LLMs. Contains datasets,test scripts,defense metrics. Implemented in Python, has 491 stars.
- AZURE AI TRAVEL AGENTS
Sample enterprise app deployed on Azure Container Apps, using MCP and multiple AI agents orchestrated by Langchain.js, LlamaIndex.TS and Microsoft Agent Framework. Agents automate flight and hotel search and build routes in real time.
- R3BL OPEN CORE TUI IN RUST
TUI framework in Rust with ready-made components:editor, file manager and process manager. Supports themes, plugins and cross-platform builds.
- RUN ANY AGENT ON ANY MODEL
llmman lets you run any AI‑agent on any model stored as an OCI image. It supports Claude Code, Codex, llama.cpp, MLX and other popular LLM frameworks.
- OKF AGENT MEMORY GIT
Git-native memory for AI agents: OKF v0.2, sub-300µs BM25 search in memory, embedded MCP server, progressive disclosure. Reduces token bloat by 80% without external databases, pure Go.
- PULLMD: CONVERT CONTENT TO MARKDOWN
Self-hosted service converts URLs and files (pages, documents, images, audio, YouTube) to Markdown. Supports PWA, REST, MCP, Claude Code skill, Reddit‑aware, and refreshable share links.
- VOICEMEM UNIVERSAL MEMORY
VoiceMem provides infrastructure for the next generation of voice agents with universal memory. The left hemisphere stores facts, the right — emotions, and the fully streaming architecture eliminates latency.
- PANE: MANAGE MULTIPLE AGENTS
Pane — open-source manager of AI agents for any CLI agent, any OS. Run locally or self‑host Remote Pane to control agents from desktop or phone via runpane CLI.
- R2AI: AI REVERSING
r2ai — plugin for radare2 that uses LLM for automatic analysis and reversing of code. Integrates directly into the radare2 interface, providing hints and generating pseudocode. Supports OpenAI models and local LLMs via API.
- KMCP: CLI AND K8s FOR MCP
kmcp is a CLI utility and Kubernetes controller that allows building, testing, and deploying MCP servers. Integrates with Go projects and supports declarative deployment in the cluster.
- GENTLEMAN PI: AGENT HARNESS
Gentle‑Pi turns Raspberry Pi into an architect harness for AI‑agents: it supports SDD/OpenSpec, subagents, strict TDD evidence and a review‑guard system, and also unlocks agent skills for developers.
- VELocUT: AI VIDEO EDITOR BROWSER
Velocut — a local-first video editor running entirely in the browser on Rust/WASM with WebGPU and WebCodecs. Users and LLM agents control it via a single JSON command protocol.
- LAUNCH AI AGENTS IN OBSIDIAN
The plugin lets you run Claude Code, Codex and other CLI agents directly inside Obsidian notes. You get an interactive terminal with AI support without leaving your workspace.
- MCP SERVER FOR CODE CONTEXT
aoci-code — local MCP server and CLI in Go. Indexes code and DB schemas into a unified repo map. Helps AI agents (Claude Code, Cursor, Codex) understand complex systems faster.
- ABC-GRPO: ALL-QUADRANT BOUNDING MODEL
ABC-GRPO implements all-quadrant bounded clipping for GRPO, improving LLM training stability. Code based on arXiv:2601.03895, written in Python. Project has 442 GitHub stars. It also ensures stable training. It works with the Python.
- BEST LIST OF AGENT SERVICES
Awesome‑agent‑native‑services — a curated list of services for AI agents: email, browsers, memory, sandboxes, payments and MCP tools. The repository includes selection criteria and step‑by‑step guides for integrating each component.
- AGENTIC RAG WITH RL SUPPORT
Harness-RL — agentic RAG framework where RL optimizes search and generation. Supports GRPO, work with LLM, and extension via Python API.
- BLOCKBENCH MCP PLUGIN
This plugin adds MCP server to Blockbench for AI tool integration. Allows managing 3D models via MCP protocol directly from the interface.
- RUOYI-WEB AI AGENT CHAT
RuoYi-Web provides a modern web interface for communicating with AI agents, configuring prompts, and working with knowledge bases via RAG. Users can ask questions, get answers, and manage agents directly in the browser.
- INTERCEPTOR: AI AGENT IN BROWSER
Interceptor provides a single, powerful CLI and MCP‑server that enables AI agents to control a signed‑in browser, native macOS applications and a genuine iPhone. It supports automation of input, clicks and scripts without extra drivers now.
- JEV TYPESAFE CATALOG
Collection of 400+ resources for structured LLM output: SDKs, agent tools, demos, and independent Jev and TypeSafe evaluations. Helps choose the best solution for typed AI responses.
- MEMSTACK: STRUCTURED FRAMEWORK CLOUD
Memstack gives 130 skills of Claude Code with persistent memory and TokenStack compression. The local dashboard launches three agents in real time, and MCP extends the AI agent's tools.
- LARADASHBOARD: FULL CMS ON LARAVEL
Laradashboard — ready-made CMS on Laravel, covering everything from basic to enterprise level. Managing users, roles, permissions, modules, settings, translations, content and monitoring. Includes Tailwind, Livewire, REST API and unit‑tests.
- 49-IDE: MANAGING AI AGENTS
Open-source 2D IDE for managing AI agents in the terminal, CLI, Git and issues. Enables working with files and issues across multiple projects and machines, self-host via localhost or Tailscale, coming soon on app.49agents.com.
- CHATGPT PLANS, CODEX EXECUTES
ChatGPT acts as the project's brain: it analyzes requirements and forms a plan. Codex turns the plan into working code, preserving context via Model Context Protocol. The result is seamless work of the AI agent without manual switching.
- VIBE-CODING TOOLKIT: AI AGENTS
vibe-coding-toolkit — a battle‑tested set of AI‑coding tools from production: Claude Code plugins, sub‑agent orchestration, quality gates, and ready‑to‑copy prompts. 409 ★, JavaScript.
- DISCIPLINED DEVELOPMENT WITH AI
Disciplined-AI-Software-Development provides an approach for AI-assisted development with constraints, validation hooks, and session management. PAG enables structured interaction with LLM. Copy claude-setup/ into your project to start today.
- PARALLEL AUDITOR OF SMART CONTRACTS
Bountyforge — a set of skills for Claude Code, launching parallel agents for auditing smart contracts (EVM, Move, Solana, TRON), web and API security. It prepares reports ready for submission to HackerOne, Bugcrowd, Intigriti and Immunefi.
- NEUROARXIV: AI CHECKS ARXIV BEFORE CODING
A skill that eliminates the need to write code from scratch: Claude searches real papers on arXiv before proposing a new architecture. This reduces the risk of reinventing the wheel and accelerates research.
- REARK AI-ANALYSIS APK AND HAP
ReArk — smart reverse engineering tool for HarmonyOS and Android. It analyzes HAP, APP, APK and AAB, using an AI-agent to output functions and build agentic-workflow. The report shows API calls, permissions and vulnerabilities.
- AI MOVIE NARRATOR
Generates voiced video recaps of movies from a single text prompt. Combines an LLM for the script, FFmpeg for editing, and Python scripts for full automation. The result is a ready‑to‑publish clip with narration and subtitles.
- COMFYUI SKILLS FOR OPENCLAW
ComfyUI_Skills_OpenClaw gives ready‑made skills for agents OpenClaw, Hermes Agent, Codex and Claude Code. They work with the local MCP ComfyUI, simplifying AI‑art pipelines. In Python, 395 stars.
- PULSE: MONITOR OF AI ASSISTANT QUOTAS
Pulse — a floating macOS monitor showing remaining quotas of Claude Code, Codex, Antigravity, OpenCode Go and Kimi Code. Written in Swift using AppKit, it updates data in real time via each service's API.
- TYPE-SAFE JEV FOR AI
Project shows evidence-backed cases, patterns, prompts and starter code for TypeSafe Jev — a System One model providing typed solutions with confidence assessment. Includes templates for classification and integration with AI‑API.
- EVIL-WINRM-PY FAST WINRM SHELL
evil-winrm-py — Python implementation of an interactive WinRM shell for remote Windows management. Supports Kerberos, NTLM and pass‑the‑hash, works over HTTP/HTTPS with arbitrary ports. Allows executing commands, uploading/downloading files and running PowerShell scripts.
- BACKGROUND AI AGENTS FOR OPCODE
Project provides background AI agents in Claude style for OpenCode, allowing async delegation with context preservation between sessions. Agents now run in background via MCP and SDK, ensuring continuous operation without manual intervention.
- CUSTOMIZATION LLM VIA LOGITSPROCESSOR
A set of LogitsProcessors for LLM behavior tuning. Solves specific tasks without changing model architecture.
- JSON REPAIR FOR LLM
json-repair fixes JSON broken by LLM models. It restores syntax, preserving structure and data types. Works in streaming mode and suits agents, function calls and embeddings.
- MCP SERVER FOR AZURE DEVOPS
MCP server for Azure DevOps gives AI agents access to work items, pipelines and repositories via MCP, enabling seamless instant interaction with tasks, PRs and builds!!
- FAUXNIX: BASH ON WINDOWS
fauxnix lets run Linux‑commands on Windows without VM or WSL by translating bash to PowerShell. It provides MCP server and CLI for AI agents (Claude Code, Codex, OpenCode), gives GNU‑style output and bash‑style errors, supports UTF‑8, GBK.!!
- VISUAL-SKILLS AI DIRECTOR FOR AGENTS
The visual-skills project gives agents AI‑director skills: cinematic dramaturgy (Murch, blocking, montage) and exact prompts for Seedance 2.5, Kling 3.0 Turbo/Omni, Veo 3.1, Nano Banana 2, GPT Image 2.5. 380★ on GitHub.
- GPT IMAGE-2 FOR E‑COMMERCE
Skill gpt‑image2‑ecommerce generates product images for e‑commerce: 39 templates, routing via GPT‑Image‑2/2.5 (Flare/Sunburst), consistency lock campaign's, and for platform compliance..
- CCTEAM ONE AGENT TEAM
ccteam unites AI agents (Claude Code, Codex, Grok, DeepSeek Harness, Kimi, Pi) into one team. A session creates tasks, distributing them among providers and collecting results on any machine; control is via Telegram, Lark or a browser.
- REMOVES ADS BEFORE PLAYBACK
MinusPod — a self-hosted Python server that intercepts media streams and removes ads before playback using LLM Claude. Users get clean content without interruptions, without needing extra plugins.
- CURATED FINTECH LIBRARIES LIST
Collection of 370+ open source libraries for payments, banking, and financial infrastructure. Saves hours of searching for ready solutions.
- AI CODE REVIEW IN PR
genai-code-review — GitHub Action that uses an LLM to analyze diff in pull request and posts comments. Just add a step to your workflow, specify the API key and choose a model.
- CC-SESSIONS-VIEWER:AI CODE ASSISTANT
cc-sessions-viewer shows sessions for cc, codex, grok, kimicode, pi, antigravity & opencode CLI. Allows managing the skills, collecting the token statistics, searching globally, resuming and exporting, to HTML.
- RUST FRAMEWORK FOR AGENTS
High‑performance Rust framework for building interactive workflow systems. Supports complex chains and multi‑agent systems with low latency.
- TEMPLATE GENAIOPS PROMPT FLOW
Template GenAIOps with Prompt Flow simplifies creation of LLM applications through centralized hosting, lifecycle management, and experiments with prompt variants and hyperparameters. Supports A/B deployment and reporting for all runs.
- BUILD LLM IN SECONDS
CLI tool creates ready-to-train PyTorch projects for LLM in seconds. Analog of create‑next‑app, but for language models. Written in TypeScript, already has 362 stars.
- GENERATE IMAGES VIA CLI
Generate images from the command line using your ChatGPT subscription — no OPENAI_API_KEY, no gateway, no daemon. Zero‑dep Python CLI ready to work as an AI‑agent skill.
- MANCODE: AI AGENT CODE
MANCODE — AI‑agent framework with five modes from practice to playoffs. It blocks over‑engineering, forcing lean, maintainable and effective codes.
- STREAMING MARKDOWN RENDER IN TERMINAL
streamdown — a Python utility that streams Markdown rendering directly in the terminal. Supports live preview and works with LLM output. Stars: 358.
- GPT‑5.6 CLAUDE GROK DEEPSEEK PROMPT WORKBENCH
ColdBrew Zero — environment for creating and testing prompts for GPT‑5.6, Claude, Grok and DeepSeek V4 Pro. Assembles instructions from «breakline» blocks, checks their effectiveness and jailbreak resistance, exports to JSON/YAML.
- FAST DATA ANALYSIS IN TYPESCRIPT
TypeScript library powered by DuckDB for tabular, geospatial, vector, AI and Google Sheets data workflows, plus visualization. Supports Deno, Node.js and Bun, delivering high‑speed analytics.
- LUVUS: MANAGING AI AGENTS
Luvus is a control panel for orchestrating AI agents, written in Rust. Start, monitor and interact with Claude, Codex and other LLMs via a unified API.
- SHELLORACLE: AI ASSISTANT TERMINAL
ShellOracle — terminal utility that generates shell commands from natural language using LLM. Supports Bash, Fish and Zsh, integrates with Gemini and other models. Speeds up work for developers and Linux administrators.
- SMART CODE READING FOR AI
tilth finds definitions, callers, and function dependencies via tree-sitter. Works without indexing — single binary for CLI and MCP server. Install: cargo install tilth or npx tilth.
- DONSETCH: WEB SEARCH FOR AI AGENTS
Donsetch is a Rust library for extracting, searching, and crawling web pages without API keys or registration. It supports anti‑bot bypass, PDF handling, and MCP protocol, licensed under AGPL v3.
- SKILLS LIBRARY HARMONYOS NEXT FOR AI
Offline skill library for HarmonyOS NEXT AI code assistants. Contains API 12‑26 declarations, ArkTS/ArkUI, NDK, DevEco Studio, emulator automation, debugging, publishing, and reproducible local environment.
- LOCAL AI ASSISTANT
Nurture AI assistant on computer. Local-first: memory, data and apps remain plain files in workspace. Chat calls right interface — wiki, spreadsheet, chart, form, 3D.
- SMALL ENGINEER ROBOT WITH AI ASSISTANT
Tiny‑Engineer — a desktop robot on ESP32 that brings an AI coding assistant to life, moving when it reads files, thinks, and writes code. After finishing a task, it demonstrates an antigravity effect, signaling readiness.
- MCP SERVER FOR YAHOO FINANCE
The MCP server provides access to Yahoo Finance data: stock prices, financial statements, options, and news. Written in Python, it allows retrieving historical prices and company information via a single API.
- MODULAR PLATFORM FOR CLAUDE CODE
Modular plugin marketplace for Claude Code and agentic CLI. Offers validated, spec‑driven skills, agents, commands, and workflows for Java, TypeScript, Python, PHP, AWS, and AI.