Coding & Development

openOutpaint is a local, offline JavaScript and HTML canvas outpainting tool designed for use with AUTOMATIC1111's Stable Diffusion webUI API. It allows users to extend images beyond their original boundaries with an effectively infinite, resizable, and scalable canvas. Key features include soft inpainting support, SDXL compatibility, a layer system, and the ability to save, load, import, and export workspaces. It also offers an inpainting/touchup mask brush, webUI script support, and a dedicated img2img tool with optional border masking. The tool is available as an extension for webUI and requires the --api flag to be enabled in the webui-user launch script.

The Alignment Handbook offers a comprehensive collection of robust, open-source recipes designed to align language models with both human and AI preferences. It addresses the growing need for resources on training these models, collecting appropriate data, and measuring performance. The handbook provides scripts for various stages of the alignment pipeline, including continued pretraining, supervised fine-tuning (SFT) for chat, and preference alignment using methods like Direct Preference Optimization (DPO) and Odds Ratio Preference Optimisation (ORPO). It also includes recipes to reproduce models like Zephyr 7B and guides on data collection and method explanations. The project supports distributed training with DeepSpeed ZeRO-3 and parameter-efficient fine-tuning with LoRA/QLoRA, making it a valuable resource for developers and researchers working on advanced LLM alignment.

Alpaca-CoT is an instruction fine-tuning (IFT) platform designed to unify interfaces for instruction collection, various large language models (LLMs), and parameter-efficient methods. It aims to simplify the use of LLMs for research and development by providing extensive instruction-tuning data, including Chain of Thought (CoT) datasets, and integrating multiple LLMs like LLaMA, ChatGLM, Bloom, and MOSS. The platform also supports various parameter-efficient methods such as LoRA, P-tuning, adalora, and prefix tuning, allowing researchers to easily switch and compare different approaches. Alpaca-CoT significantly improves CoT reasoning and the ability to follow Chinese instructions, making it a valuable resource for NLP researchers looking to reduce the threshold for getting started with large language models.

AnyGPT is an open-source, unified multimodal large language model (LLM) that leverages discrete representations for processing diverse modalities, including speech, text, images, and music. The base model aligns these four modalities, facilitating seamless intermodal conversions between them and text. It also features the AnyInstruct dataset, built from various generative models, which provides instructions for arbitrary modal interconversion. This allows the chat model to engage in free multimodal conversations, where different data types can be inserted at will. AnyGPT employs a generative training scheme that converts all modal data into a unified discrete representation, utilizing the Next Token Prediction task for unified training on an LLM. This approach aims to compress vast amounts of multimodal data into a single model, potentially unlocking capabilities not found in pure text-based LLMs.

Are-copilots-local-yet is a comprehensive, open-source resource that tracks the current trends and state-of-the-art in local LLM copilots. It focuses on tools for code completion, project generation, shell assistance, and automatic bug fixing, enabling developers to find solutions that run on consumer machines. The platform highlights the benefits of local copilots, such as offline and private use, improved responsiveness, better project context awareness, and the ability to run specialized models. It provides curated lists of editor extensions, tools, chat interfaces, and relevant models and datasets, making it an invaluable guide for developers interested in the frontier of local AI-powered development.

awesome-ai-tools is a meticulously curated list of Artificial Intelligence tools, designed to help users discover and explore the rapidly evolving landscape of AI. The repository features a wide array of tools, including top generative AI tools, large language models (LLMs), AI tools for developers, marketing, and various other applications. It categorizes tools for easy navigation, covering areas like AI text generation, code with AI, generative AI for images, video, and audio, as well as AI phone call agents and learning resources. The platform encourages community contributions, allowing users to submit their AI tools for free and stay updated with regular additions and updates through the Altern Newsletter.

Awesome-AITools is a comprehensive GitHub repository that curates a wide array of AI-related utilities, making it a valuable resource for developers, researchers, and AI enthusiasts. The collection spans various categories, including AI chatbots like Claude, Gemini, and ChatGPT, as well as open-source LLMs such as Llama 3 and Mixtral. It also features AI agents like Auto-GPT and Claude Code, along with tools for AI coding, image creation, video creation, and speech processing. The repository encourages community contributions through pull requests, ensuring a continuously updated and expanding list of practical AI tools. It's designed to help users efficiently discover and leverage the latest advancements in AI technology across different applications.

GPTs Today is an AI tool that, based on its name and prior description, aims to help users discover and explore various GPTs. The platform was designed to feature a comprehensive list, allowing users to search and filter based on specific criteria. Key functionalities previously included knowledge integration, DALL·E image generation, code interpretation, and web browsing capabilities. However, the current live website content indicates that the site is redirecting, suggesting it may be undergoing maintenance, a redesign, or is no longer actively hosted at the provided URL. Therefore, its current operational status and features cannot be confirmed from the live site.

PingCRM is an AI-powered, open-source, and self-hostable personal networking CRM designed to help professionals maintain and strengthen their valuable relationships. This tool unifies interactions from platforms like Gmail, Telegram, Twitter/X, and LinkedIn into a single timeline per contact, providing a comprehensive view of communication history. Leveraging AI, PingCRM identifies relationships that are "slipping away" and proactively drafts contextual follow-up messages using models like Claude, making it effortless to re-engage. It features relationship scoring, identity resolution, and a weekly digest to surface important contacts. Ideal for anyone whose success depends on a robust professional network, it ensures timely attention to important contacts.

awesome-artificial-intelligence is a comprehensive, curated list of actively maintained resources for building and shipping AI systems. It focuses on AI engineering, including Retrieval-Augmented Generation (RAG), AI agents, evaluation frameworks, guardrails, and deployment strategies. The resource list features a selection of modern and practical books, guides, and playbooks for AI engineering, such as 'Designing Machine Learning Systems' and the 'OpenAI Cookbook'. It also highlights landmark papers that shaped modern AI, like 'Attention Is All You Need', and provides structured content through courses from institutions like Stanford and Google. Additionally, it includes newsletters to stay current with AI developments and a variety of tools for building and deploying AI applications, from models like ChatGPT and Claude to developer tools like GitHub Copilot and multimedia AI tools for image, video, and audio generation.

HyperLLM offers a solution for developers and researchers working with language models, focusing on efficiency and cost reduction. The platform provides small language models specifically designed for fine-tuning and training, with the goal of significantly lowering the associated costs by up to 85%. This tool emphasizes instant fine-tuning capabilities, making it a valuable resource for those who need to quickly adapt models to specific tasks or datasets. It caters to AI researchers and machine learning engineers looking for optimized and cost-effective ways to develop and deploy custom language models.

Stable-Diffusion is an extensive open-source project offering expert-level tutorials and resources for generative AI. Led by Dr. Furkan Gözükara, it covers a wide array of topics including FLUX, Stable Diffusion, SDXL, SD3, LoRA, Fine Tuning, DreamBooth, and various WebUIs like Automatic1111, Forge, and ComfyUI. The project provides guides for advanced techniques such as deepfake generation, text-to-video, animation, and voice cloning. It also includes practical tutorials for using platforms like Google Colab, RunPod, and Kaggle, making it a comprehensive resource for both learning and developing generative AI applications.

Interactive 3D and 2D Visualization of GPT-2 offers a unique and engaging way to understand the complex architecture of large language models. This tool provides both 2D and 3D visualizations of the GPT-2 (124 M) model, allowing users to delve into its internal mechanisms. Users can explore layers, attention scores, and token processing in real-time, with options to skip through animation passes and view equations. It includes features like KV cache mode for prefill and decode, and a dev mode for vector sampling. This platform is an invaluable educational resource for anyone looking to demystify how LLMs process information and generate text.

speech-swift is an open-source Swift library designed for on-device speech AI on Apple Silicon. It provides advanced speech recognition (ASR) models like Qwen3-ASR and Parakeet TDT, which surpass OpenAI's Whisper Large v3 in accuracy and speed, running up to 40x real-time. The library also includes text-to-speech (TTS) with voice cloning, speech-to-speech, voice activity detection (VAD), speaker diarization, and speech enhancement. All functionalities operate locally on Apple devices, leveraging MLX for GPU inference and CoreML for the Neural Engine, ensuring privacy and efficiency without cloud dependencies or API calls.

EnLume Inc. specializes in empowering innovators and disruptors through well-crafted IT solutions, focusing on AI, Machine Learning, and Data services. They assist businesses in integrating advanced AI and ML algorithms into their operations to drive automation and facilitate data-driven decision-making. EnLume builds robust data pipelines and develops AI-powered analytics platforms, ensuring efficient data processing and insightful analysis. Their offerings include cloud-agnostic solutions, providing flexibility and scalability across various cloud environments, and a strong emphasis on product engineering excellence to deliver high-quality, reliable IT solutions tailored to client needs.

Tracerny is a robust prompt injection defense solution designed to safeguard AI applications from malicious attacks. It provides a free SDK with 258 real-world attack patterns, enabling local detection with sub-5ms latency and no data collection. For enhanced security, the Pro version includes Layer 2 LLM Sentinel for output validation, context-aware scanning, and delimiter salting. Tracerny works with any LLM, including GPT-4, Claude, Gemini, and Llama, and is production-ready, used by companies handling sensitive data. Its two-layer defense model ensures comprehensive protection by stopping suspicious inputs and validating LLM outputs.

Awesome-Agentic-Reasoning is a comprehensive, open-source repository that curates papers and resources focused on agentic reasoning for large language models. It systematically organizes research into thematic areas such as planning, tool use, search, self-evolution through memory and feedback, multi-agent systems, and real-world applications. This resource is based on the survey "Agentic Reasoning for Large Language Models" and aims to bridge the gap between thought and action in autonomous agents. It's an invaluable tool for researchers and developers looking to stay updated on the latest advancements, offering insights into foundational, self-evolving, and collective reasoning paradigms, as well as core mechanisms and diverse applications.

Geeky Bee AI Private Limited is an artificial intelligence company focused on providing development services in cutting-edge fields such as computer vision, deep learning, and automation. The company is dedicated to tackling complex challenges for a global clientele, leveraging advanced technologies to deliver innovative and effective solutions. Geeky Bee AI emphasizes affordability, making high-end AI capabilities accessible to a broader market. Their expertise spans various AI domains, enabling them to build custom solutions tailored to specific business needs, from intelligent automation systems to sophisticated image and video analysis tools. They position themselves as a partner for businesses looking to integrate AI into their operations to enhance efficiency and drive innovation.

Modulate is a frontier voice AI company specializing in Ensemble Listening Models (ELM), which are designed to outperform traditional LLMs in understanding real conversations. Their core product, Velma, is an ELM that analyzes how something is said, not just what, providing nuanced insights into emotion, intent, and context. Modulate offers APIs for transcription, deepfake detection, and upcoming voice analytics, catering to enterprises and developers. The platform helps businesses detect voice fraud and deepfakes, improve customer experience in contact centers, and monitor AI agents. Modulate emphasizes high accuracy and cost-effectiveness, with Velma ranking #1 in various benchmarks for conversation understanding, transcription, and deepfake detection.

distil labs provides a platform for training and deploying custom small language models (SLMs) that are designed to be faster, cheaper, and as accurate as larger LLMs. The platform automates the fine-tuning process by creating synthetic data from production traces and then training a model for specific tasks. Users can upload traces, train their models with a single command, and deploy them to hosted or local endpoints. The deployed models are OpenAI-compatible, allowing for seamless integration into existing workflows. distil labs supports various text processing tasks including classification, QA, and tool calling, and offers significant cost savings on inference.

agent-shell is an Emacs buffer designed for seamless interaction with LLM agents powered by the Agent Client Protocol (ACP). It allows users to chat with a variety of agents, including Gemini CLI, Claude Agent, Auggie, and Mistral Vibe, all within a native Emacs environment. The tool supports extensive configuration for authentication with different providers like Anthropic, Google, and OpenAI, including API keys, OAuth tokens, and login-based methods. It also enables passing environment variables to spawned agent processes and loading them from .env files. agent-shell is highly extensible, with additional packages available for features like Claude Agent skills, mobile interaction via Slack, sandboxed AI coding agents, and dedicated workspace management.

ImageBind by Meta is an advanced AI model designed to integrate and understand information across six different modalities: images, videos, audio, text, depth, and thermal data. This multimodal approach allows the model to create a unified representation of various sensory inputs, enabling more comprehensive AI understanding and interaction. It supports conversions between different media types, such as generating audio from an image or creating an image from text, opening up new possibilities for creative applications. ImageBind is particularly useful for developing interactive narratives, enhancing AI performance in recognition tasks, and exploring novel ways to combine diverse data streams for richer AI experiences.

Maincode is an AI research and product company based in Melbourne, Australia, dedicated to building advanced AI solutions. Their flagship product, Matilda, is an intelligent AI assistant designed to understand complex workflows, reason about context, and take meaningful action. Unlike typical chatbots, Matilda aims to learn user patterns, perform multi-step reasoning, and utilize agentic tool use to get work done. Maincode's research areas include a long-context reasoning framework, an action layer for tool-use and agentic execution, and an internal evaluation suite for model capability and safety. They emphasize building AI that understands context, takes action, and earns trust through reliable results.

awesome-ml is a comprehensive, curated list of resources designed for professionals and enthusiasts in the fields of Large Language Models (LLM), analytics, and data science. This GitHub repository serves as a central hub for discovering open LLM models, development tools, and various AI-related assets. It covers a wide array of topics including native and web GUIs, backends, voice assistants, retrieval augmented generation, browser extensions, and AI agents. Additionally, the list delves into multimodal AI, code generation libraries, prompt templating, fine-tuning, model merging, and quantization. Researchers and developers will find valuable sections on datasets, research papers, product showcases, benchmarking, leaderboards, and optimization techniques, making it an indispensable resource for staying updated in the rapidly evolving AI landscape.