Trending AI projects

MIRA is a real-time world model of Rocket League that uses a 5B parameter latent diffusion model to simulate 2v2 matches based on player actions.

Google AI Edge LiteRT Samples is a genuine open‑source collection of on‑device machine‑learning demos, model‑conversion recipes, and automation tools for LiteRT (the successor to TensorFlow Lite) and LiteRT‑LM. It provides Android, iOS, and Web/WASM examples covering speech, vision, text‑to‑image, and multimodal apps, plus scripts to convert Hugging Face checkpoints into LiteRT‑compatible bundles. The repo is aimed at developers building high‑performance edge AI applications.

Zero-WAM is a causal video-action policy that enables robots to perform unseen tasks zero-shot by using human demonstration videos or language as in-context specifications.

RepoPrompt CE is an open‑source macOS app that helps you collect and review the exact parts of a codebase (files, diffs, CodeMaps) that an LLM‑based coding agent needs, then hands that curated context to the agent via a built‑in MCP server. It also lets you launch and coordinate CLI coding agents from the UI, supporting multi‑repo workspaces and token‑budget‑aware prompt building.

MyContext is a desktop app that continuously ingests a user’s work data (chat, docs, meetings, etc.) into a private, locally stored knowledge graph. The graph can be queried by LLMs, local agents, or authorized AI apps, providing them with a personal context layer for more informed responses. Features include a digital‑self assistant, natural‑language search with source citations, and strict human‑in‑the‑loop controls. The project is actively developed and released under the Elastic License 2.0.

dstack is a unified control plane that orchestrates GPU resources across cloud providers, Kubernetes, and on-premise clusters to streamline AI development, training, and inference.

A lightweight LLM scheduling framework that reduces prefill overhead by reusing KV caches for frequently accessed external knowledge across LLM systems.

A real-time physics simulation engine and SDK that provides tools for simulating physical interactions, destruction, and fluids, including Python bindings for USD physics.

SayIt is an open‑source Windows app that lets you dictate anywhere via a global shortcut, then uses speech‑to‑text and optional LLM‑based cleanup to insert polished text at the cursor. It supports local GGUF ASR models, cloud APIs, or a self‑hosted FastAPI backend, offers context‑aware editing, per‑app rules, overlay feedback, and full history. Licensed under AGPL‑3.0.

AAS is an open‑source, Docker‑based framework for multi‑drone development, fast ROS 2 simulation (PX4/ArduPilot, Gazebo, YOLO/LiDAR), and deployment on Jetson hardware, with optional RL Gymnasium support.

A lightweight, standalone C++ inference engine for Google's Gemma models, designed for research and experimentation with a minimalist, CPU-optimized implementation.

A family of high-precision end-to-end speech recognition models optimized for noisy environments, far-field audio, and extensive Chinese dialect support.

Silero Models is an open‑source collection of neural text‑to‑speech models (Russian, Turkic, Caucasian, and other CIS languages). Install via `pip install silero` or load from PyTorch Hub, then synthesize speech with a single `model.apply_tts(text)` call. The models are fast on CPU/GPU, support multiple sampling rates, and include language‑specific features like automated stress for Russian and SSML markup for fine‑grained control. Licensed under CC BY‑NC 4.0 (with MIT‑licensed CIS base variants).

A GitHub repository that curates system prompts from various AI agents and provides a guide analyzing them to extract best practices for designing effective prompts for agentic AI systems.

A practical Russian-language course on Claude Code, from first run to professional team setup, with 12 modules, 8 labs, a starter kit, prompt recipes, and automation examples for CI and GitHub Actions.

Bops is an open‑source macOS app that lets you run a team of AI bots to automate business tasks. Bots execute on real cloud VMs (Orgo), can be chatted with in‑app or via Slack/Telegram/Discord, email, and phone, and integrate with ~1,400 apps. Free, Pro, and Max plans provide cloud compute and AI credit; the repo also supports full self‑hosting with your own API keys.

A self-improving context engine for coding agents that uses Markdown documents to maintain durable, shareable memory of project specs and decisions across sessions.

Intent is a desktop app that coordinates AI coding agents, notes, tasks, git and terminals via a Rust backend daemon (`intentd`). All state lives locally in SQLite, accessed through JSON‑RPC over a Unix socket or secure WebSocket. The packaged app bundles the daemon; advanced users can install the daemon separately. It supports multiple AI providers, optional GitHub/Linear/Sentry integrations, and an upcoming iOS companion. No telemetry is sent; updates are pulled from GitHub releases only.

Metis is an open-source coding agent that runs as a desktop app, terminal tool, or Skill, letting each project define its own agent architecture. It automates planning, coding, review, and testing through editable workflows and automatic task tiers.

Docker Skills is a set of portable, standards‑compliant *Agent Skills* that teach AI coding assistants how to handle Docker tasks—Dockerfiles, Compose stacks, sandbox lifecycles, and Docker Agent configuration—while providing safety guardrails.

An AI-native tutoring system implemented as a Claude Code skill that provides adaptive, personalized courses on AI design and Forward Deployed Engineering.

DeepMind’s representations4d provides research‑grade video models that learn 4‑D (spatial + temporal) features via masked auto‑encoding, moving‑token attention, recurrent masking, multimodal alignment, and diffusion‑based perception. It ships pretrained checkpoints (up to 22 B parameters), Colab demos, and a small Python library for inference on depth, tracking, segmentation, and text‑prompted multi‑task vision.

DeepJIT is a header‑only C++20 library that provides just‑in‑time compilation and caching of CUDA or Huawei Ascend kernels, with PyTorch‑friendly Python bindings and shared‑cache support.

Show‑Harness is an open‑source framework that lets a vision‑language model control robots via a tiny token‑based action vocabulary. It supports zero‑shot use of large VLMs and fine‑tuned small VLMs, works with multiple robot platforms and simulators, provides a browser UI (GUMI) for collecting demonstrations, and includes a plugin system for perception/planning extensions. Released LoRA adapters and datasets are hosted on HuggingFace.

An MCP-based plugin for Unreal Engine that exposes over 1,400 editor actions to AI assistants using a namespace-dispatch pattern to minimize context window usage.

Marigold V2 is an open‑source research codebase that converts a frozen diffusion transformer (Qwen‑Image‑Edit‑2509) into fast, single‑step dense predictors for monocular depth, see‑through depth, surface normals, and albedo. It provides pretrained LoRA checkpoints, a lightweight fine‑tuning recipe (≈ 1 week on a single 32 GB GPU), CLI inference, benchmark evaluation scripts, and a modular YAML‑driven training framework that can be extended to new dense‑vision tasks.

A context engineering tool that converts entire codebases into structured LLM prompts, automating the process of gathering and formatting code for AI models.

A full-system DNN hardware exploration platform that allows architects to design and evaluate the performance of AI accelerators using a configurable systolic array architecture.

gawkbot is an open‑source macOS/iPhone app that turns natural‑language workflow descriptions into dedicated, locally‑run AI bots. Each bot has its own UI, schedule, tools, data tables, and a knowledge wiki with citations. Bots use Claude Code, Codex, Opencode, or any local model (Ollama, Hermes, OpenClaw) and can act on 1 200+ integrations, but every external write is paused behind a manual approval gate. The app lives in the Mac’s camera notch, offers keyboard/voice shortcuts, supports multi‑workspace, team sharing (Tailscale or Cloudflare tunnel), and stores all state in local files. It’s free, self‑hosted, and aims to provide a privacy‑first, cost‑transparent alternative to cloud‑only “always‑on” AI assistants.

An agentic AI assistant for the Linux CLI designed to help users learn and master ethical hacking and cybersecurity.

A collection of packaged instructions and scripts that extend AI coding agents' ability to understand and work with Firebase.

homr is an open‑source Python OMR tool that uses a UNet segmentation model and a transformer‑based symbol recognizer to turn photos or PDFs of sheet music into MusicXML files, runnable via a simple CLI on CPU or GPU.

A curriculum-aligned knowledge graph and dataset for STEM subjects, designed to benchmark and train LLMs to understand the structured prerequisites and relationships of K-12 education.

ComfyUI‑Pixaroma adds a large set of user‑friendly nodes to the ComfyUI visual workflow engine, covering image/video/3D I/O, prompt handling, notes, logic controls, resizing, and built‑in editors. It streamlines common tasks, offers extensive documentation, and is installed via Easy‑Install or a simple Git clone.

An official MCP server that connects AI assistants and IDEs to Perplexity's real-time search, reasoning, and research capabilities.

ToolUniverse is an open‑source platform that lets any LLM act as an AI scientist. It provides a standardized protocol, a catalog of 1 000+ scientific tools, a local MCP server, a CLI (`tu`), and a Python SDK for discovering, composing, and running tools—enabling agents like TxAgent and Medea to perform drug‑discovery and omics analyses.

A desktop AI agent that operates within your local environment, automating tasks across files, terminals, browsers, and office software.

An automated comic translator that uses LLMs, OCR, and inpainting to translate comics and manga from multiple languages while preserving the original visual layout.

An open-source, AI-powered presentation generator that creates customizable slides from a topic via an outline-first workflow.

An open-source inference engine that consolidates multiple open-source models for agent tasks into a single self-hosted cluster with a unified API.

SAHI is a Python library that adds tiled (sliced) inference to existing object‑detection models, enabling accurate detection of small objects in very large images or videos. It supports many popular detectors, offers a CLI and FiftyOne integration, and includes COCO‑dataset utilities.

mcpc is a universal command‑line client for the Model Context Protocol (MCP). It maps every MCP operation—tools, prompts, resources, tasks, skills, OAuth authentication, and even experimental x402 payments—into intuitive shell commands. Designed for AI agents and developers, it lets you inspect, debug, and script MCP workflows using standard Unix tools (e.g., jq, xargs). Features include persistent named sessions, progressive tool discovery via `grep`, JSON‑only output mode, secure OAuth 2.1 storage (OS keychain or protected file), AI sandboxing, and lightweight cross‑platform support.

A platform for building AI agents featuring a no-code visual builder and a TypeScript SDK with full two-way synchronization for collaborative agent management.

A developer-oriented platform for the full lifecycle management of AI agents, providing tools for prompt engineering, automated evaluation, and execution observability.

An AI Gateway and LLM observability platform that provides unified API access to 100+ models, request tracing, and prompt management for AI engineers.

SAG is a local-first knowledge base application and retrieval architecture that replaces traditional RAG and GraphRAG with event-entity indexing and query-time dynamic hyperedges. It turns uploaded documents into searchable, connected, traceable knowledge with cited agent answers, and claims state-of-the-art retrieval and QA results on HotpotQA and other benchmarks.

Vespa is an open‑source platform for large‑scale search, recommendation and AI inference. It stores text, structured data, vectors and tensors, runs machine‑learned models at query time, and delivers results with sub‑100 ms latency on distributed clusters.

cocoindex‑code is a real, open‑source tool that builds an AST‑based semantic index of a codebase and lets you search it with natural‑language queries. Install via pipx (full or slim variant), run `ccc init` and `ccc index`, then use `ccc search` or integrate the provided `ccc` skill/MCP server with Claude, Grok, Oh My Pi, Codex, etc. Docker images are also available for reproducible deployments.