Trending AI projects

helm-d is a DeepSeek Harness plugin that enforces evidence-based security analysis—every conclusion must cite disk-based E-numbered evidence, and it ships 34 tools, 362 on-demand references, and hot-swappable jailbreak profiles for Android, Web, Native, Protocol, Malware, and AI-Security tasks.

StreamEval is an open-source benchmarking framework designed to evaluate the performance of interactive streaming video generation models across long-duration sessions. It provides standardized tasks and metrics to measure how well these models maintain context and respond to live user inputs.

WeVisOne-VL is a Tencent vision-language model that unifies object-level tasks (detection, counting, pointing, OCR, referring) and pixel-level tasks (segmentation, matting) in one system, using a 9.6M-parameter pixel decoder and structured text tokens. It comes in 2B and 4B sizes, reliably rejects absent targets, and outperforms open baselines across all headline benchmarks. Inference code, demos, and evaluation pipelines are released; fine-tuning code is still pending.

Roast Room is a Node.js web app that uses Claude Code sub‑agents to simulate a council of seven famous business figures. You feed a decision, the agents argue, score options, and produce a verdict PDF—all driven by curated “brains” built from each person’s public work.

Reason Network is a peer‑to‑peer daemon that lets AI coding assistants (Claude, Codex, Cursor, Gemini, etc.) on different developers’ machines discover each other, share status, and exchange messages instantly, all without a central server.

A low-level orchestration framework for building, managing, and deploying long-running, stateful AI agents with durable execution and human-in-the-loop capabilities.

A framework for building agents and LLM-powered applications that simplifies development through interoperable components and a vast library of third-party integrations.

A unified API gateway providing direct, proxy-free access to multiple frontier LLMs like GPT, DeepSeek, and Claude using a standardized OpenAI protocol.

A free, on‑premise Python SDK for face detection, landmark extraction, embedding generation and similarity scoring, targeting Windows/Linux applications that need privacy‑preserving biometric matching.

ARS‑Codex is a Codex‑compatible skill that bundles the Academic Research Skills workflow suite (originally built for Claude Code) into a single plug‑in. After installing the plugin, users can invoke `$academic‑research‑suite` and let Codex route the request to one of six research‑oriented workflows—deep research, paper drafting, reviewer simulation, full research‑to‑paper pipeline, systematic‑review screener, or experiment planning. The repo contains the skill manifest, prompt templates, optional runtime adapters, and installation scripts; it does not include a new model.

ARIS is a workflow and toolkit for running long, autonomous research tasks with AI coding agents. It uses a two-model loop: an executor agent performs the work, and an independent reviewer model audits each step. It provides skills, a CLI, and plugins for multiple agent platforms, enabling reliable 'set-and-forget' research.

A tool that connects AI agents to Chrome or Edge, allowing them to operate within the user's logged-in browser sessions to perform web tasks and debug websites.

daily_stock_analysis 是一个基于大语言模型的全自动股票分析系统，支持多市场数据聚合、AI 生成的买卖报告、Web 工作台、Agent 策略问股以及通过企业微信、飞书、Telegram、Discord、Slack、邮件等渠道的自动推送。提供 GitHub Actions 一键部署或本地/Docker 部署方式，使用 MIT 许可证。

A library of 818 structured cybersecurity skills and playbooks designed to give AI agents the domain knowledge and decision-making workflows of a senior security analyst.

PI‑Desktop is a cross‑platform, local‑first desktop application that provides a persistent, plugin‑extensible workspace for AI agents. It lets you connect any LLM (cloud or local), orchestrate multiple agents or sub‑agents, and build custom functionality via plugins (panels, commands, widgets, services, etc.). All data stays on your machine, and the system is built on the open‑source pi agent ecosystem.

An open-source, self-hosted AI agent that automates the entire YouTube content lifecycle—from research and script writing to video production and analytics—while requiring human approval for every upload.

NiubiGEO is an open‑source, self‑hosted web app that lets you query LLMs about a domain (your product), records the raw answers, citations, competitor mentions and keywords, and lets you repeat or schedule tests. It supports multiple model providers, shows evidence for every answer, and offers a keyword‑monitoring module for ongoing brand visibility tracking.

Stop That Shit is a Node‑based plugin/skill that guards AI coding agents (Codex, Claude Code, OpenCode, Hermes, Pi, etc.) against four wasteful behaviours—scope creep, unnecessary hashing, intent violations, and task thrashing. It analyses diffs, intercepts file writes, dependency installs, sub‑agent spawns, and checksum generation, blocking them unless the user explicitly allows the action. Installable via each host’s plugin system, it provides simple commands (`review`, `change`, `monitor`, …) to enforce the guard, with an optional “STSS” skill for trimming defensive prose. MIT‑licensed.

Franco is an outbound sales agent that runs on your own machine and uses your existing Claude Code or Codex subscription instead of a model API bill. It finds or imports accounts, writes or sends approved email sequences, reviews replies, and records every fact with its source.

MaiBot is an open‑source, QQ‑based chatbot built on large language models. It focuses on human‑like conversation, memory of individual users, and a plug‑in system for extensibility. The project provides a one‑click launcher, detailed docs, and an active community, and is released under GPL‑3.0.

The repository is a genuine AI‑powered patent‑assistant built with AgentSkills and Playwright, offering multiple LLM‑driven skills for drafting disclosures, applications, analysis, and more.

Pi-Bolt is a high-performance, ahead-of-time compiled version of the Pi coding agent that runs as a native executable, significantly reducing startup time and resource usage without requiring Node.js or Bun.

A tool for detecting Claude running environment risks and account ban risks by analyzing browser fingerprints, IP purity, and system settings.

Dashi PPT Skill is a Node‑based plugin that lets AI agents generate, edit, and export fully‑styled slide decks. It offers 12 themes, 1 020 templates, rich chart/model layouts, in‑browser controls, and one‑click PPTX/PDF export, all running locally under an AGPL‑3.0 license.

A collection of AI-agent skills and motion-graphic templates for automating video editing tasks like cutting, captioning, and beat-syncing within coding environments.

LinkedIn‑agent‑skill is a set of Claude prompt‑skills plus two tiny Python scripts that generate, clean, and score LinkedIn posts, comments, profile rewrites, weekly plans, carousel slides, DMs, inbox triage, and audits. It never posts automatically; you copy‑paste the final draft. The humanizer removes invisible characters, normalises punctuation, swaps buzzwords, and runs five local heuristics to give a “human‑score”. Install by copying the `li‑*` folders into `~/.claude/skills/` and optionally filling a `voice.md` file with your writing style.

AIRI is an open‑source AI‑driven virtual companion that can chat, play games, and animate a VRM/Live2D avatar. It runs on browsers, desktop (Electron) and mobile (PWA), supports many LLM providers, speech‑to‑text, multi‑provider TTS, and GPU‑accelerated graphics via WebGPU, CUDA or Metal.

PandaCode is an open‑source native desktop app that lets you run many AI coding agents (Claude Code, Codex, Kimi Code, etc.) alongside Git, a terminal and a browser. It offers a Kanban task board, agent‑team orchestration, bot integrations, local/cloud sandboxes, one‑click deploys and deep Git support, all stored locally with no telemetry.

A tool that identifies the underlying LLM behind an API by analyzing the unique statistical patterns in the numbers the model generates, helping users detect unauthorized model swapping.

An open-source resume builder and job application tracker that features ATS checks and optional AI-powered writing assistance.

Raven is a host agent harness that orchestrates multiple specialized agents (research, code, design, oncall) for complex projects, and includes a runtime self-improvement loop where a Curator rewrites its own strategy modules. It also connects third-party agents and has been shown to autonomously deliver complete games, research projects, and product launches.

AI破甲 is a Codex configuration and Skill installer that sets up a custom instruction file and Skill to guide AI in authorized tasks, with verification and recovery tools. It is based on a modified version of gpt-instruct v45, but explicitly removes certain aggressive behaviors. It does not claim universal compatibility or proven demo success.

MiniMind is a lightweight, from-scratch LLM framework that enables individuals to train and experiment with language models on consumer hardware for minimal cost. It provides a complete educational pipeline covering pre-training, fine-tuning, and reinforcement learning without relying on complex third-party abstractions.

An extremely fast Python package and project manager written in Rust that replaces pip, poetry, and pyenv to streamline dependency and version management.

Skill Store is a curated, automatically synced catalog of AI‑agent “skills” (JSON‑described plugins) covering document handling, content creation, video production, e‑commerce automation, and more. It pulls 182 official skills from the community‑maintained awesome‑agent‑skills repo every 24 h and adds 63 hand‑picked Chinese‑language skills. The repo provides Python scripts to update, export, and validate the skill definitions, plus a web UI for browsing. Ideal for developers building LLM‑orchestrated agents who want ready‑made, reusable capabilities.

A Claude skill that generates ready-to-use, token-efficient prompts for ~30 different AI tools (LLMs, coding agents, image/video/voice generators, automation platforms), using a pipeline of intent extraction, clarifying questions, template selection, and anti-pattern fixes.

A Claude Code skill that guides you through interactive Q&A, remembers your preferences, and automatically writes a full Chinese novel using LLMs, with options for serial or parallel agent writing and built‑in validation.

Grandet is an open‑source skill that lets AI agents look up the cheapest listed API price for any LLM model, discover free‑credit offers, and generate the exact configuration needed to connect the chosen endpoint. It works via a website, a streamable MCP server, or direct installation into agents like Claude Code, Cursor, and others.

Skills Manager is a cross‑platform desktop and CLI tool for collecting, organising, and deploying AI‑agent skill packages (prompt‑code bundles) across many coding assistants. It offers a unified library, marketplace, presets, global/project workspaces, multi‑tool sync, Git‑backed backup, and agent‑aware management, all built with React/Tauri (frontend) and Rust/SQLite (backend).

A component-based WebGPU library that allows developers to integrate complex visual effects into web apps via a declarative API, visual editor, and AI-assisted CLI tools.

Sno Station Skills is a collection of 37 professional habits and workflows installed into AI agents to enable peer reviews, automated task handoffs, and self-improving loops. It solves the problem of unreliable AI output by enforcing rigorous, team-like engineering standards.

A tool that converts SRT subtitles into whiteboard-style animation videos by synchronizing hand-drawn elements with narrative timing.

MemPalace is a local-first memory system for AI assistants (Claude Code, Codex, Cursor, any MCP client). It stores conversation histories verbatim — no summarization — and retrieves them via semantic search from a structured index organized into "wings" (people/projects), "rooms" (topics), and "drawers" (content). Pluggable vector backends (ChromaDB default; SQLite-exact, Rust-native, Milvus, Qdrant, pgvector optional) keep everything on-device by default, with no API calls needed. It ships a CLI, a 47-tool MCP server, a temporal knowledge graph, auto-save hooks, and Docker images. Claims 96.6% raw retrieval recall (R@5) on LongMemEval without any LLM, with reproducible benchmarks and unusually honest methodology notes.

A searchable index of 16,844 open-source projects that exist because someone reverse-engineered a closed app, API, device, protocol, file format, game, or AI product. Includes a website, an AI-agent skill, and MCP server, refreshed nightly, with health status per project and an explicit scope that excludes exploits, bypasses, and DRM-stripping tools.

A model-definition framework for state-of-the-art machine learning that provides a unified API for inference and training across text, vision, audio, and multimodal models.

DeerFlow is an open‑source framework that lets you build autonomous AI agents capable of orchestrating sub‑agents, using tools (web search, sandboxed code execution, RAG), and managing memory. It supports many LLM providers, offers a web UI, a terminal UI, and integrates tracing services, all deployable via Docker.

LingBot‑Map is a feed‑forward transformer that reconstructs 3‑D point clouds from streaming video in real‑time, using a paged KV‑cache (FlashInfer) and optional sky‑masking. It provides interactive and offline demo scripts, pretrained checkpoints, and supports very long sequences via windowed inference.

dsh‑infinite‑gen‑5 is a Windows‑only DeepSeek Harness plugin that adds up to 38 AI‑driven reverse‑engineering tools—ranging from decompilation and interactive CFG visualisation to Unicorn‑based emulation, x64dbg debugging, crypto‑analysis and protocol parsing. Tools are split into a lightweight Core 8 default set and an optional Full 38 set, with write‑capable actions gated by a public approval service. Installation is a one‑click PowerShell script that validates a fixed asset bundle; the plugin is operated via slash commands inside the DSH chat UI and includes seven pre‑packaged analysis runbooks. The project is released under CC BY‑NC‑SA 4.0 with strict non‑commercial and legal usage restrictions.