Trending AI projects

SREGym is an AI-native benchmarking platform that provides live Kubernetes environments to evaluate and train AI agents in diagnosing and mitigating real-world Site Reliability Engineering (SRE) failures.

SuperSonic is an AI-powered business intelligence platform that combines natural language querying with a semantic data layer to provide reliable, governed data insights.

SentrySearch is a command-line tool that indexes video footage into a local vector database, letting you search with natural-language text or a reference image and save the best matching moment as a trimmed clip.

UI-Venus-2 is a unified AI agent for controlling mobile, web, and desktop GUIs. It improves on version 1.5 with more mobile environments, desktop support, keypoint-based verification, and reflection learning. The README provides quick-start scripts for mobile, computer, browser, grounding, and CAPTCHA tasks, plus two frameworks for real-device automation. Benchmark tables show strong performance, especially the 27B model, on mobile and computer tasks.

TapCanvas is an open‑source, AI‑agent‑driven platform that lets creators build scripts, storyboards, images, videos and subtitles on a typed “infinite canvas”. Agents read the canvas, plan multi‑step production workflows, call text/image/video models via a unified gateway, and store every asset with provenance. The monorepo includes a React UI, Hono API, DeepSeek Harness bridge, and a model‑gateway (Lu‑Ban) that supports many providers (HeyRoute, Factory/Droid, etc.). Docker‑compose boots the whole stack; the core code is MIT‑licensed.

ArcadeDB is a high-performance multi-model database that supports graph, document, key-value, time-series, vector, and geospatial data with multiple query languages and an MCP server for AI integration. It runs embedded or remotely via various drivers, offering ACID transactions and 70+ graph algorithms.

A framework for building federated AI systems that enables decentralized model training across various ML frameworks and devices.

A set of plain-text rules and instructions that enable AI agents to write in a clear, concise style and remove unnecessary complexity from their responses.

An open-source quadruped robotics framework for ESP32-based boards, enabling affordable gait research, reinforcement learning, and physical AI development.

MinisSkills is a community collection of reusable skills for Claude, each packaged as a SKILL.md file that extends the assistant with specialized workflows, domain knowledge, and optional bundled resources like scripts and references. It solves the problem of giving Claude domain-specific expertise on demand through a standardized skill format.

A hands-on, project-based course (in a GitHub repo) that teaches Python developers to build real AI agents with LangChain and LangGraph — spanning 10 projects from a simple hello-world to self-correcting agents, with instructions to follow git commits as lessons.

ai-cookbook is a practical set of AI code recipes and tutorials for developers who want to copy, paste, and integrate AI functionality into their own projects.

LaRecipe is a Laravel package that hosts your documentation inside your app, provides a beautiful reader-facing site, and integrates with coding agents so they update the docs automatically with every code change. It also offers Markdown export, llms.txt, and an MCP server for agents.

This project is the companion resource for a 66-page survey paper on "harness self-evolution" — the process by which an AI agent improves the software around its model (prompts, memory, workflows, tools) using past execution experience, without retraining the model itself. It provides a definition, taxonomy, categorized literature list, related surveys, evaluation and theory branches, and background precursors.

A project that builds and evaluates n-gram and neural language models to rank Chinese pinyin candidate readings for input methods. It ships a shared evaluation set and two tiny Transformer models (4.5 MB and 25.5 MB) that match or beat much larger open models in quality while being dramatically faster and lighter.

AI-FW is a self-hosted gateway that sits between AI agents and the models they call, inspecting every prompt and response, enforcing governance rules, and producing regulator-ready audit logs. It supports fail-closed operation, agent identity, cost-saving caches, and integration with tools like Claude Code, Cursor, and M365 Copilot.

A self-hostable, open-source workspace that integrates chat, task management, and knowledge base tools with AI agents, featuring governed approval workflows and support for the Model Context Protocol.

ComfyUI‑H3‑FaceRefine adds custom nodes that detect, crop, re‑generate, and stitch back tiny faces in MiniMax H3 videos, fixing the model’s poor performance on small heads. It offers automatic and manual face‑selection workflows, robust tracking, optional cut detection, and integrates with existing MiniMax H3 pipelines.

An open-source Optical Music Recognition (OMR) application that transcribes music score images into symbolic digital formats for playback and editing.

A fast, language-agnostic code navigator that indexes codebases into SQLite to enable millisecond-latency symbol lookups, dependency tracing, and impact analysis for humans and AI agents.

gflow-cli is a Python CLI (and optional MCP server) that automates Google Flow’s Veo video‑generation and Imagen image‑generation via a real Chrome session. It supports batch jobs, reusable characters, prompt‑rewriting tools, and can pipe results to cloud storage. Designed for developers and AI‑agent integration, it runs in alpha and requires a headed Playwright‑Chromium browser.

A local observability tool for AI coding agents that visualizes project progress and activity through a structural map or a 3D kitchen-themed interface.

A tool and field guide designed to identify and remove generic, AI-generated design patterns (AI slop) from web projects.

Crabbox is a CLI that lets you lease temporary computers (cloud VMs, local containers, or any SSH host), sync your repo (including uncommitted changes), run commands/tests, stream output, and release the box. It supports many providers, warm‑up reuse, desktop/VNC access, and an optional coordinator for shared credential and spend management—ideal for AI coding agents and cross‑platform testing.

torchtitan is a PyTorch‑native library for large‑scale LLM pre‑training, fine‑tuning, and reinforcement‑learning, offering composable parallelism (FSDP, TP, PP, CP, EP), recipe‑based configuration, and support for frontier models and low‑precision formats.

Pipelock is an open‑source Go proxy that sits between AI agents and the internet, scans outbound/inbound HTTP, WebSocket, and MCP traffic for secrets, prompt‑injection, SSRF, and other attacks, and emits cryptographically signed receipts proving each allow/deny decision. It offers strict, balanced, and audit modes, a free offline evidence viewer, and an enterprise dashboard with RBAC and fleet monitoring.

A tool that enables DLSS 5 neural rendering in games that lack native DLSS support by simulating the necessary API calls using ReShade data.

A high-performance Rust-based web research tool for AI agents that provides stealthy fetching, keyless search, and crawling without requiring API keys.

AgentDock is an open‑source Go application that exposes a secure HTTP‑based Managed Control Plane (MCP) on each computer or server. It lets LLM chat assistants (ChatGPT, Claude, Codex, etc.) perform file operations, run shell commands, manage Git, automate browsers and desktop apps, and coordinate work across multiple devices—all from a single AI conversation.

A high-performance, geo-distributed federated metadata lake that provides unified access and governance for data and AI assets across multiple sources and regions.

A Simplified Chinese rewriting skill that removes stilted, formulaic AI-generated patterns to make text sound natural while preserving all original facts and technical terms.

A library of 187 installable AI agent skills for academic research, featuring a specialized suite for Nature-style scientific writing and manuscript preparation.

LingBot‑VLA 2.0 is an open‑source Vision‑Language‑Action foundation model for robots. It unifies many robot embodiments into a 55‑dim action vector, uses MoE layers and dual‑query distillation (depth + future video) to learn predictive dynamics, and is trained on ~60 k h of robot and egocentric video data. The repo provides conda‑based installation, scripts to download 6 B‑parameter checkpoints, tools for data preparation, post‑training on custom datasets, and evaluation on the RoboTwin benchmark and real robots. Reported results show noticeable gains over the 1.0 version on bimanual and mobile manipulation tasks.

ChatGPT2API is an open‑source self‑hosted service that turns the ChatGPT web UI into an OpenAI‑compatible REST API, with a Vue admin console for multi‑account, proxy, quota and asset management.

LiteReality‑Agent is an open‑source pipeline that converts phone‑captured RGB‑D scans (via the LiteReality iOS app) into fully articulated, physics‑ready 3‑D indoor scenes. It cleans layout, infers object parts, assigns masses/inertias, generates textures with LLM‑driven image APIs, and exports to MuJoCo URDF/MJCF, Blender, or GLB. The heavy AI models run on Modal’s free GPU service, so even Apple‑silicon Macs can use it. One CLI command (`uv run litereality run <scan>`) performs the whole reconstruction, after a brief `uv`‑based environment setup.

LogoCreator is an open‑source web app that generates brand‑ready logos using the FLUX‑2 image model via Together AI. Built with Next.js, TypeScript, Radix, and Tailwind, it runs without an account—just a Together AI key. Features include instant generation, FLUX‑1 edits, PNG/SVG export, optional rate‑limiting and auth, and a history dashboard. The repo provides clear setup steps and lists future tasks such as size selection, cost preview, reference‑logo uploads, and a showcase of redesigned famous logos.

A collaboration tool for AI engineers and domain experts to build and curate high-quality datasets for NLP, LLMs, and multimodal models.

A scalable platform for running AI agent environments using Firecracker microVMs, featuring rapid snapshot-based booting and high-density resource management.

An implementation of the Model Context Protocol for the Unity Editor, allowing AI assistants to directly manipulate scenes, assets, and editor settings.

MuseBot is a chatbot that connects eight popular messaging platforms (Telegram, Discord, Slack, Lark, DingTalk, WeChat Work, QQ, and WeChat) to large language models such as OpenAI, DeepSeek, and Gemini, enabling AI-powered responses with support for images, documents, voice, function calling, and RAG. It can be deployed locally or via Docker in about a minute.

GuideLLM is a Python library (also available as a container) that benchmarks LLM inference against real‑world workloads. It drives configurable traffic patterns to OpenAI‑compatible or vLLM servers, supports synthetic and real multimodal datasets, records detailed latency/token metrics, and outputs JSON/CSV/HTML reports for capacity planning and SLO verification.

DeepSWE is a benchmark for measuring coding agents on 113 long-horizon software engineering tasks from real-world open-source repositories.

TokenHub is an open‑source, Go‑based gateway that adds enterprise‑level governance (routing, permissions, budgeting, and bill reconciliation) to any OpenAI‑compatible LLM provider. It offers role‑aware consoles, OpenAI‑compatible APIs, SQLite/PostgreSQL deployment options, and Docker‑Compose or systemd installers.

K-Dense BYOK is an open-source AI research assistant that runs locally, helping scientists analyze data, review literature, and generate reports while maintaining a transparent, verifiable lab notebook.

Newsjack is an open‑source toolkit that turns LLM agents into a PR team. It provides a set of Markdown‑based “skills” (e.g., angle generation, fact‑checking, journalist finding) and a small CLI, installable via a one‑line script or as a Claude.ai plugin. The skills run fully on local agents (Claude Code, Codex, Hermes, OpenClaw) and in limited one‑shot mode on Claude.ai and ChatGPT. It targets founders, agencies, and marketers who want AI‑assisted news‑jacking, monitoring, and PR workflow automation.

An MCP server providing fast, cheap, typed judgment tools from TypeSafe's Jev model to help AI agents perform high-frequency verification and screening tasks.

Brigade is a self‑hosted, multi‑agent AI assistant framework. It lets you run a crew of agents that share persistent, provenance‑aware memory (Tideline), are organized in an org‑chart, and can be reached via terminal UI, WhatsApp, Slack, Discord, iMessage, etc. All API keys and data stay on your machine; you can bring any LLM (Claude, OpenAI, Gemini, local Ollama, etc.) and connect to 1 000+ third‑party apps through Composio. Features include skills, sub‑agent fan‑out, cron scheduling, document/media processing, optional Convex storage, and an MCP memory server. Install with a one‑line script or via npm, then onboard, start the always‑on gateway, and chat.

zerostack is a lightweight coding agent written in Rust that runs in the terminal, supporting multiple LLM providers. It offers granular permission controls, session management, subagents, and optional persistent memory, while using far less RAM and CPU than JavaScript-based agents.