AI Trending

Project directory

Projects

Open-source AI projects tracked by the leaderboard, ordered by GitHub stars.

Go 88 stars

Lifailon/openrouter-bot

This project allows to launch your Telegram bot in a few minutes to communicate with free or paid AI models via OpenRouter.

Overview: This project allows developers to quickly set up and launch a Telegram bot that communicates with free or paid AI models through OpenRouter, or local language models like LM Studio. Key features: - Launching the bot in minutes - Communication with free and paid AI models via OpenRouter - Local language models via LM Studio - Easy integration with Telegram bot creation tools Tech stack: Go Why it matters: This project streamlines the process of creating and deploying a Telegram bot, making it accessible and efficient for developers, especially those working with AI models.
Python 88 stars

arsbr/Veritensor

The Anti-Virus for AI Artifacts & RAG Firewall. A static analysis tool scanning Models and Notebooks for RCE, Datasets and RAG docs for Data Poisoning, PII, and Prompt Injections. Secure your AI Supply Chain.

### Overview: Veritensor is the Anti-Virus for AI Artifacts and the ultimate Firewall for RAG pipelines. It secures the entire AI Supply Chain by scanning the artifacts that traditional SAST tools miss: Models, Datasets, RAG Documents, and Notebooks. ### Key Features: - **Native RAG Security:** Veritensor integrates seamlessly with `LangChain`, `LlamaIndex`, `ChromaDB`, and `Unstructured.io` to block threats at runtime. - **High-Performance Parallel Scanning:** Utilizes all CPU cores with robust SQLite Caching (WAL mode). Re-scanning a 100GB dataset takes milliseconds if files haven't changed. - **Advanced Stealth Detection:** Detects prompt injections using CSS (`font-size: 0`, `color: white`) and HTML comments. - **Dataset Security:** Scans Parquet, CSV, JSONL, and Excel datasets for "Poisoning" patterns and malicious URLs. - **Archive Inspection:** Safely scans inside `.zip`, `.tar.gz`, and `.whl` files without extracting them to disk. - **Dependency Audit:** Checks `pyproject.toml`, `poetry.lock`, and `Pipfile.lock` for malicious packages and vulnerabilities. - **Data Provenance:** Generates a signed JSON snapshot of your data artifacts for compliance (EU AI Act). - **Identity Verification:** Verifies model hashes against the official Hugging Face registry to detect Man-in-the-Middle attacks. - **De-obfuscation Engine:** Automatically detects and decodes Base64
JavaScript 88 stars

bennyschmidt/ragdoll

The library for character-driven AI experiences.

Overview: Ragdoll is a library designed to create character-driven AI experiences, offering versatile knowledge sources and distinct personalizations. Key features: - Knowledge scraping from provided URLs - Generation of personalized responses - Extensibility for various deployment methods Tech stack: JavaScript Why it matters: Ragdoll simplifies the implementation of AI characters in games and applications, ensuring that chatbot interactions feel more natural and relevant to the context. Repository: bennyschmidt/ragdoll GitHub stars: 88
HTML 88 stars

braviaprime/proctorai

AI‑powered web exam system with face recognition, webcam, browser, and audio checks ensures integrity in remote assessments. Flask prototype reached ~72% functional accuracy and 90%+ detection rates, reducing cheating in online education.

Overview: ProctorAI is an AI-powered web exam system designed to enhance the integrity of remote assessments through real-time behavioral monitoring. Leveraging computer vision, audio analysis, and context-aware logging, it detects disruptive behaviors and unauthorized activities to prevent cheating. Key features: - Real-Time Video Anomaly Detection - Audio Signal Processing - Context-Aware Logging - Explainable Outputs Tech stack: Python, Flask, OpenCV, WebRTC, Oracle Cloud Infrastructure (OCI) Free Tier. Why it matters: ProctorAI aims to reduce false positives and provide context-aware anomaly detection, advancing Trustworthy and Explainable AI in education.
Python 88 stars

nMaroulis/protolink

Build autonomous Python agents with native Agent-to-Agent (A2A) communication

### Overview: ProtoLink is a lightweight, **A2A**-first Python framework designed for building **pluggable agents** and multi-agent systems. It leverages the **A2A** architectural core, treating each agent as a self-contained runtime entity with identity, capabilities, lifecycle, tools, and direct task-based communication. ### Key Features: - **A2A Communication**: Native `AgentCard`, `Task`, `Message`, `Part`, and `Artifact` runtime model built on A2A 0.3. - **LLM-agnostic and Local-first**: Provider-native tool calling when available, strict JSON fallback for self-hosted models. - **Pluggable Modules**: Agents can be composed of optional modules such as APIs, local LLMs, RAG, tools, transports, registries, storages, state, telemetry, authentication, logging, policy, and durable run records. - **Simple by Default, Explicit When Needed**: Agents are stable compositions, and only what is needed is plugged in. ### Tech Stack: - **Python**: Primary language. - **Pydantic**: Used for data validation and models. - **Pydantic**: Used for data validation and models. - **Ruff**: Used for code formatting and linting. - **Ty**: Used for code formatting and linting. - **Ask DeepWiki**: Used for asking AI questions. ### Why It Matters: ProtoLink simplifies the development of multi-agent systems by providing a pluggable architecture that allows agents to be composed of optional modules. This approach keeps the
Python 88 stars

nestyme/Subtitles-generator

generates transcript for video from link

Overview: This project generates subtitles for videos from a given link using Google Speech API and YouTube DL. Key features: - Automatically generates subtitles for videos uploaded to YouTube. - Supports downloading videos from any link. - Uses Google Speech API for speech recognition. Tech stack: Python Why it matters: This tool streamlines the process of creating subtitles for videos, making it a valuable resource for anyone looking to automate the transcription process.
Python 88 stars

niopeng/HyperRIM

Overview: HyperRIM is a PyTorch implementation of a conditional deep generative model designed to avoid mode collapse and generate multiple outputs for the same input, increasing the width and height of images by a factor of 16x, and recovering plausible images from badly compressed ones. Key features: - Increased image resolution - Mode collapse prevention - Multiple output generation for the same input Tech stack: PyTorch Why it matters: HyperRIM is a significant advancement in generative modeling, offering practical solutions for image augmentation and image recovery, making it valuable for applications requiring high-resolution images or image restoration tasks. GitHub stars: 88
Rust 88 stars

nnethercott/hannoy

Production-ready KV-backed HNSW implementation in Rust using LMDB

Overview: Hannoy is a production-ready key-value backed HNSW implementation in Rust utilizing LMDB as a storage backend. It supports various metrics and is designed for environments with limited memory. Key features: - Supports metrics including Euclidean, Cosine, Manhattan, and Hamming distances. - Provides Python bindings with maturin and pyo3. - Enables multithreaded builds using rayon. - Uses disk-backed storage for indexing datasets that exceed memory limits. - Utilizes compressed bitmaps to store graph edges efficiently. - Supports dynamic document insertions and deletions without full re-indexing. Tech stack: Rust Why it matters: Hannoy is designed to be efficient and scalable, making it suitable for environments where memory is a constraint, such as multi-programming setups or cases where the dataset size exceeds available RAM. Its disk-backed storage and non-blocking concurrent reads ensure reliability and performance in multi-threaded environments. GitHub stars: 88
TypeScript 88 stars

pixeltable/pixelbot

Multimodal AI agent, an interactive data studio with on-demand ML inference, media generation, and a database explore

Overview: Pixelbot 3.0 is a local multimodal AI application powered by Pixeltable, offering on-demand ML inference, media generation, and a database explorer. It integrates with Pixeltable's declarative schema, query routes, and FastAPI API, ensuring efficient storage and retrieval of various media types. Key features: - Pixeltable integration for efficient data storage and retrieval - On-demand ML inference capabilities - Media generation and processing - Database exploration and analysis Tech stack: TypeScript Why it matters: Pixelbot 3.0 simplifies the integration of AI and data management, making it easier to build interactive data studios and applications that leverage Pixeltable's powerful capabilities. Repository: pixeltable/pixelbot GitHub stars: 88 GitHub description: Multimodal AI agent, an interactive data studio with on-demand ML inference, media generation, and a database explore
Python 88 stars

rollroyces/deepcatch

DeepCatch: performance-weighted multi-modal fusion for ultra-early cfDNA cancer detection. Fragmentomics, methylation, sequence.

### Overview: DeepCatch is an open-source computational framework designed for ultra-early cancer detection using cell-free DNA (cfDNA). It fuses complementary molecular modalities through a self-supervised Transformer model, tracks patients longitudinally with Bayesian Kalman filtering, and predicts tissue-of-origin in a two-stage Capture → Enhance → Triage pipeline. ### Key features: - **Multi-modal Fusion**: DeepCatch fuses fragmentomics, methylation, and sequence data to improve detection accuracy. - **Bayesian Kalman Filtering**: Tracks patients longitudinally to enhance prediction accuracy. - **Tissue-of-Origin Prediction**: Predicts the tissue of origin for cfDNA samples. ### Tech stack: Python 3.9+ ### Why it matters: DeepCatch is a research tool designed for benchmarking and research purposes. It provides a standardized benchmark for evaluating early cancer detection methods using cfDNA. The framework is open-source, allowing researchers to reproduce and build upon the results.
Go 88 stars

sanonone/kektordb

AI memory system combining vector search with temporal knowledge graph. Built-in cognitive engine for agents. Supports memory decay, contradiction detection, and MCP integration.

### Overview: KektorDB is an AI memory system that combines vector search with a temporal knowledge graph, providing persistent memory for agents. It includes a cognitive engine that continuously analyzes and updates data, detects contradictions, and builds user profiles. ### Key Features: - **Cognitive Engine**: Analyzes and updates data, detects contradictions, and builds user profiles. - **Graph Operations**: Supports CRUD, graph operations, session management, and cognitive features. - **Memory CRUD**: CRUD operations for memories. - **Graph Operations**: Connects entities, finds paths, and explores connections. - **Session Management**: Starts and ends sessions. - **Cognitive Features**: Checks subconscious, gets user profile, asks meta questions. - **Knowledge Compilation**: Compiles structured knowledge. - **Indexing**: Provides indexing capabilities. - **Configuration**: Configures the system. ### Tech Stack: Go ### Why it Matters: KektorDB is essential for agents that need persistent memory, continuous data analysis, and intelligent decision-making based on user interactions and knowledge graph updates.
Swift 88 stars

sawfwair/mere-run

Run local image, text, speech, vision, music, and video workflows, plus model management and a loopback OpenAI-compatible API server. The repo ships the core libraries and the mere.run executable.

### Overview: Run local image, text, speech, vision, music, and video workflows, plus model management and a loopback OpenAI-compatible API server. The repo ships the core libraries and the mere.run executable. ### Key features: - Local image generation - Text processing and code generation - Speech synthesis and text-to-speech - Vision and image analysis - Music generation and synthesis - Video creation and editing - Model management and offline guide - OpenAI-compatible API server ### Tech stack: Swift ### Why it matters: The mere.run project provides a powerful and flexible local AI runtime for developers and creators. It allows for seamless integration of AI into local workflows, making it ideal for use cases requiring quick and easy access to AI capabilities without the need for cloud services.
Jupyter Notebook 88 stars

sayakpaul/dreambooth-keras

Implementation of DreamBooth in KerasCV and TensorFlow.

### Overview: This repository provides an implementation of DreamBooth using KerasCV and TensorFlow. DreamBooth is a method for quickly fine-tuning Stable Diffusion on new visual concepts. The code is based on Hugging Face's `diffusers` example and is intended for research purposes only. ### Key Features: - Implementation of DreamBooth using KerasCV and TensorFlow. - Codebase tested with `sks` as the unique identifier and `dog` as the class. - Two types of prompts: instance prompt and class prompt. - Instance images: 3-10 representative images of the concept to be fine-tuned. - Class images: 200-300 images generated using the class prompt for prior-preservation loss regularization. ### Tech Stack: KerasCV and TensorFlow ### Why It Matters: DreamBooth is a technique for quickly fine-tuning Stable Diffusion on new visual concepts. The code provided in this repository is for research purposes only. It is intended to help researchers and developers understand and experiment with the DreamBooth method.
Unknown 88 stars

speech-translator-ext/speech-translator-readme

This repository outlines the procedures and general information for the Speech Translator project.

### Overview: This repository outlines the procedures and general information for the **Speech Translator** project. ### Key Features: - Real-time speech transcription and translation - Transcribing and translating livestreams, videos, calls, etc. - Transcribing and translating speech for a livestream overlay in OBS (for livestreamers) - Real-time computer-assisted translation - Practicing language learning by dictating the text and reading the translation - Recording transcripts and translated captions for livestreams, calls, videos, or podcasts - Extracting a text transcript to translate it later manually or with an advanced LLM ### Tech Stack: Unknown ### Why it Matters: This web browser extension can translate any video, audio, or livestream in real-time, making it useful for transcribing and translating various media sources.
C 88 stars

voicute/onnx-wakeword

Open-source, fully offline wake word / keyword spotting (KWS) inference engine — ONNX/TFLite <130KB, multi-keyword, multilingual. Train a custom wake word online with auto hard-negative mining; deploy on ESP32 / Raspberry Pi / Android / Web / Windows / macOS / Linux / Home Assistant — runs on any ONNX/TFLite-capable device.

### Overview: onnx-wakeword is an open-source, fully offline inference engine designed for wake-word detection and keyword spotting (KWS). It supports offline inference, no audio uploads, and models under 130KB. The engine is optimized for ESP32, Android, Raspberry Pi, and various web platforms, including browser, desktop, and Home Assistant. The system is self-developed and includes a training pipeline for custom keyword models. ### Key Features: - **Sub-130KB Models**: 25K parameters, fitting ESP32 INT8 flash. - **Multi-keyword Detection**: Detects 2-10+ keywords with a single model. - **Multi-language Support**: Supports Chinese, English, Japanese, French, German, Spanish, and Hindi. - **Offline Inference**: Runs completely offline, no audio uploads, no API calls, no telemetry. - **Auto Hard-Negative Mining**: One-click auto hard-negative mining improves false trigger detection by up to 99%. - **Training Pipeline**: Fully self-developed training pipeline for custom keyword models. ### Tech Stack: - **Python**: Web and Android runtime. - **TFLite**: Optimized INT8 runtime for ESP32-S3. ### Why It Matters: - **Privacy-First Design**: Ensures no audio is uploaded, no API calls, and no telemetry. - **100% Offline Inference**: Runs completely offline, ensuring audio data never leaves the device during inference. - **Zero Cost**: Supports 7 languages, including Chinese
MDX 88 stars

weaviate/weaviate-io

Website for the Weaviate vector database

Overview: Weaviate is a vector database project, and this repository serves as its documentation website. Key features: - Docusaurus 2 for static website generation - Node.js and Yarn for project management - Upstream repository forked and cloned - Tracking branch set for pull requests - Local and remote repository setup - Installation of Node.js and Yarn - Starting the yarn server for local testing - Building the web site for static content Tech stack: Node.js, Yarn, Docusaurus 2 Why it matters: This documentation website is crucial for developers to understand and contribute to the Weaviate project, ensuring clear and up-to-date information.
Python 88 stars

yoyolicoris/music-spectrogram-diffusion-pytorch

### Overview: This project provides a PyTorch implementation of the music synthesis technique described in the paper "Multi-instrument Music Synthesis with Spectrogram Diffusion". The code includes training scripts for various diffusion models, pre-trained models, and evaluation metrics, as well as inference tools for synthesizing audio from MIDI files. ### Key features: - Training scripts for different diffusion models (small without context, small with context, base with context). - Pre-trained models for VGGish and TRILL embeddings. - Evaluation metrics for reconstruction and FAD (Forward Attention Distance) across the test dataset. - Inference tool for synthesizing audio from MIDI files. ### Tech stack: PyTorch ### Why it matters: This project aims to increase the reproducibility of the paper's work by providing accessible training code and pre-trained models, making it easier for researchers and developers to replicate the results and build upon the findings.
Python 87 stars

FareedKhan-dev/Detect-AI-text-Easily

Some words that LLM regularly uses

### Overview: This project aims to detect AI-generated text by analyzing common words and phrases that appear frequently in AI-generated content, such as "delve," "intricacies," and "unwavering." By identifying these patterns, users can better recognize AI-generated text and avoid falling into the trap of AI-generated content. ### Key Features: - **Word Detection:** The project uses a corpus of over 19 billion English words to identify common AI-generated words and phrases. - **Visualization:** Graphs and charts are used to visualize the trends of these words over time, helping users understand the frequency and patterns of AI-generated text. - **Comparison:** The project compares AI-generated text with human-written content to highlight the differences and provide a critical thinking exercise. ### Tech Stack: - **Python:** The project is built using Python, a popular language for data analysis and machine learning. ### Why it Matters: - **Critical Thinking:** Recognizing AI-generated text is crucial for maintaining ethical use of AI and avoiding plagiarism. - **Educational Tool:** This project serves as an educational tool for students and educators to understand the nuances of AI-generated text and improve their critical thinking skills. ### Repository: The repository is hosted on GitHub and has been starred by 87 developers, indicating its popularity and usefulness in the community. ### Primary Language: The primary language used in the project is Python, making it accessible to a wide range of developers and researchers.
Pascal 87 stars

MaxiDonkey/DelphiGenAI

The GenAI API wrapper for Delphi seamlessly integrates OpenAI’s latest models (gpt-5 serie), delivering robust support for agent chats/responses, text generation, vision, audio analysis, JSON configuration, web search, and asynchronous operations. Image generation with gpt-image-2. Skills and agents

Overview: Delphi GenAI is a powerful wrapper for OpenAI's GPT-5 series models, designed to seamlessly integrate with Delphi applications. It supports various features such as agent chats, text generation, vision, audio analysis, JSON configuration, web search, and asynchronous operations. Delphi GenAI is optimized for Delphi 10.4, 11, and 12, and supports all platforms. Key features: - Streamed responses - Non-streamed responses - Support for OpenAI, LM Studio, and Google Gemini models - Robust integration with Delphi applications - Easy to use and integrate with existing Delphi projects Tech stack: Delphi 10.4, 11, and 12 Why it matters: Delphi GenAI simplifies the integration of advanced AI capabilities into Delphi applications, providing a powerful toolset for developers looking to enhance their applications with natural language processing, vision, and more. Repository: MaxiDonkey/DelphiGenAI GitHub stars: 86
Python 87 stars

NimbleBoxAI/nbox

The official python package for NimbleBox. Exposes all APIs as CLIs and contains modules to make ML 🌸

Overview: Nbox is an official Python package designed for interacting with NimbleBox's infrastructure and services, providing a CLI and modules for machine learning applications. Key features: - Exposes all NimbleBox APIs as Command Line Interfaces (CLI) - Supports writing and executing code in Python - Enables documentation of code with mathematical equations - Allows importing and sharing notebooks - Integrates with popular libraries like PyTorch, TensorFlow, Keras, and OpenCV - Enables sharing and collaborating on projects - Provides an intuitive dashboard for managing jobs and instances Tech stack: Python Why it matters: Nbox simplifies interaction with NimbleBox's services, making it easier to develop and manage machine learning projects efficiently. Repository: NimbleBoxAI/nbox GitHub stars: 87 GitHub description: The official python package for NimbleBox. Exposes all APIs as CLIs and contains modules to make ML 🌸