AI Agents & Automation

Humble AI is an AI management platform designed to help organizations create, manage, and share AI tools securely and privately. It integrates seamlessly with existing business applications such as Slack, your browser, Airtable, Notion, and HubSpot, eliminating the need for heavy setup or extra logins. The platform focuses on automating repetitive tasks like chasing files, drafting follow-ups, and searching for information, freeing up teams to concentrate on work requiring human judgment. Humble AI differentiates itself by providing AI that understands company policies, finds specific documents quickly, speaks in the brand's voice, and comprehends customer processes, moving beyond generic AI responses. It offers solutions for sales, marketing, customer support, HR, and operations, ensuring data privacy and compliance with standards like GDPR.

The Heurist Agent Framework is a flexible, multi-interface AI agent framework designed for building sophisticated agents with advanced capabilities. It supports reasoning, extensive tool use, memory management, and deep research functionalities. The framework facilitates blockchain interaction and enables agents-as-a-service models. It boasts a modular architecture, allowing agents to process text and voice, generate media, interact across platforms like Telegram, Discord, Twitter, and Farcaster, and store information in knowledge bases. A key differentiator is its integration with Heurist Mesh, a Web3 skills marketplace providing specialized AI agents for crypto analytics, optimized for agent input/output formats and accessible via REST API or MCP.

LazyLLM is a low-code development tool designed for building multi-agent large language model applications. It enables developers to create complex AI applications with ease and supports continuous iterative optimization. The platform offers a convenient workflow for application building, providing standard processes and tools for various stages of development. LazyLLM emphasizes rapid prototyping, data feedback, and iterative optimization, allowing users to quickly build prototypes, analyze bad cases, and fine-tune models to improve performance. It supports one-click deployment for multi-agent applications, cross-platform compatibility across bare-metal servers, Slurm clusters, and public clouds, and a unified user experience for both online and locally deployed models. Additionally, LazyLLM facilitates efficient model fine-tuning and integrates common RAG components.

LocalAI is a versatile open-source AI engine designed to run a wide array of AI models, from large language models (LLMs) to vision, voice, image, and video models, on virtually any hardware, including CPU-only setups. It boasts impressive compatibility with APIs like OpenAI and Anthropic, making it a flexible alternative for developers. The platform supports over 36 backends, including llama.cpp, vLLM, and transformers, and offers hardware acceleration for NVIDIA, AMD, Intel, and Apple Silicon. Key features include multi-user support with API key authentication, built-in AI agents for autonomous tasks, and a privacy-first approach ensuring data remains within your infrastructure. LocalAI also provides capabilities for text generation, audio processing, image generation, and real-time APIs, making it a comprehensive solution for local AI inference.

LLMLingua is a powerful tool designed to optimize the performance of Large Language Models (LLMs) by significantly compressing prompts and KV-caches. This innovative approach, detailed in research papers presented at EMNLP'23 and ACL'24, allows for up to 20x compression while maintaining minimal performance degradation. The tool helps overcome common LLM challenges such as token limits, high API costs, and the 'lost in the middle' issue for long contexts. LLMLingua offers various versions, including LongLLMLingua for enhanced long-context processing and LLMLingua-2 for faster, task-agnostic compression. It also features SecurityLingua, a safety guardrail model for detecting and mitigating jailbreak attacks through security-aware prompt compression. The tool is integrated into popular frameworks like LangChain and LlamaIndex, making it accessible for developers working on LLM-based applications.

MiroFlow is a high-performance, modular, and fully open-source framework designed for building intelligent AI agents. It excels in multi-step internet research and complex challenges like future event prediction, achieving top rankings on multiple benchmarks including FutureX, GAIA, HLE, xBench-DeepSearch, and BrowserComp. The framework features advanced multi-turn conversation capabilities, extensive tool ecosystem integration, and hierarchical sub-agent orchestration. Built with robust concurrency management and fault-tolerant design, MiroFlow efficiently handles rate-limited APIs and unstable networks, ensuring reliable execution of complex tasks. It is cost-effective, powered by the open-source MiroThinker model, and can run a research agent service on a single RTX 4090, relying on free, open-source tools for simple deployment and scaling.

NNPACK is an acceleration package specifically designed to optimize neural network computations on multi-core CPUs. It focuses on delivering high-performance implementations of convolutional neural network (convnet) layers. The tool is not intended for direct use by machine learning researchers but rather provides low-level performance primitives that are leveraged by leading deep learning frameworks such as PyTorch, Caffe2, MXNet, and Darknet. It supports various platforms including Linux, macOS, Android, and iOS, and offers multiple algorithms for convolutional layers, including Fourier transform, Winograd transform, and implicit matrix-matrix multiplication. Implemented in C99 and Python, NNPACK features multi-threaded SIMD-aware implementations and extensive unit test coverage.

OAT (Online Alignment Toolkit) is a simple yet efficient open-source framework designed for running online LLM alignment algorithms. It features a distributed Actor-Learner-Oracle architecture optimized for high efficiency, utilizing vLLM for accelerated response sampling and DeepSpeed ZeRO for memory efficiency. OAT simplifies the experimental pipeline by providing an online Oracle for preference data labeling and real-time model evaluation. Researchers can simulate various feedback types, including verifiable rewards and LLM-as-a-judge, with flexible deployment options for reward models. Its modular structure facilitates rapid prototyping and experimentation, implementing cutting-edge algorithms like PPO/Dr.GRPO for online RL and Online DPO/SimPO/IPO for preference learning, fostering innovation and fair benchmarking.

npcpy is a comprehensive Python library designed for research and development in NLP, multimodal LLMs, Agents, ML, and Knowledge Graphs. It offers a flexible agent framework that supports both local and cloud providers, enabling users to build sophisticated AI applications. Key capabilities include multi-agent team orchestration, tool calling, and advanced functionalities like image, audio, and video generation. The library also facilitates knowledge graph integration and fine-tuning of models, making it a versatile solution for developers and researchers working with diverse AI technologies. It provides quick examples for creating personas, direct LLM calls, agent with tools, streaming, JSON output, and Pydantic structured output.

Seahorse AI Agents is a software company dedicated to the design, development, and delivery of on-demand AI agents. The company specializes in creating custom AI agents tailored to specific tasks and business needs. By focusing on bespoke solutions, Seahorse aims to provide businesses and developers with the tools to implement advanced AI capabilities efficiently. Their expertise lies in transforming complex requirements into functional AI agents, offering a comprehensive service from initial concept to final deployment. This approach ensures that clients receive highly specialized AI solutions that integrate seamlessly into their existing operations.

Skills.sh is an open agent skills ecosystem that allows users to discover and install reusable capabilities, known as 'skills,' for their AI agents. These skills provide procedural knowledge, enabling agents to accomplish specific tasks more effectively. The platform features a leaderboard that ranks skills based on anonymous telemetry data, showcasing the most popular and useful skills in the ecosystem. Users can install skills using a simple CLI command, integrating them directly into their AI agents. The platform emphasizes security through routine audits to assess skills for malicious content, while also encouraging users to review skills before installation. Skills.sh aims to enhance AI agents by providing a wide array of specialized functionalities.

Pegasi is an applied AI safety and compliance layer designed for enterprise agent deployments. It helps organizations confidently deploy self-improving AI agents by providing essential guardrails and control mechanisms. The platform supports human approval workflows, ensuring that AI agent actions align with organizational policies and human oversight. Pegasi also offers comprehensive audit logs, crucial for maintaining compliance with standards like SOC 2 and HIPAA, and for providing transparency into agent behavior. This makes it an ideal solution for enterprises looking to manage the risks associated with AI agent use while ensuring reliable and secure performance.

Oraichain Labs is a pioneering AI Layer 1 blockchain platform established in 2020, offering comprehensive frameworks and tools for integrating human-centric artificial intelligence with decentralized infrastructures. The platform is dedicated to advancing AI blockchain oracle technology and fostering cross-chain interoperability, paving the way for the mass adoption of next-generation Web3 applications. Its dynamic ecosystem supports a wide range of products across DeFi, NFTs, Identity, Collective Intelligence, Asset Tokenization, and Smart Healthcare. Oraichain provides resources for developers, including technical support, business development aid, and funding for innovative ideas, making it a robust environment for building and scaling AI-driven decentralized solutions.

ROMA (Recursive-Open-Meta-Agent) is an open-source framework designed to simplify the creation of hierarchical, high-performance multi-agent systems. It employs a recursive plan-execute loop, breaking down complex tasks into parallelizable components for efficient problem-solving. Key features include an Atomizer to determine task atomicity, a Planner for subtask decomposition, Executors for handling atomic tasks, and an Aggregator to synthesize results. ROMA supports various LLM providers, offers built-in toolkits like Calculator and File operations, and provides flexible installation options from a minimal setup for quick evaluation to a full Dockerized production environment with persistence, observability, and a REST API.

Solana-agent-kit is an open-source toolkit designed to bridge AI agents with Solana blockchain protocols. It allows any AI agent, regardless of the underlying model, to autonomously perform a wide array of Solana actions. These capabilities span over 60 distinct operations, including deploying and trading tokens, launching new tokens, lending assets, and sending compressed airdrops. The kit offers extensive features for token operations, NFT management, and DeFi integration with platforms like Jupiter Exchange, Raydium, and Orca. It also includes AI integration features such as LangChain and Vercel AI SDK support, autonomous modes, and built-in error handling, making it a comprehensive solution for developers building AI-powered applications on Solana.

sktime is a comprehensive open-source Python library designed for machine learning with time series data. It offers a unified interface for various time series learning tasks, including forecasting, time series classification, clustering, anomaly detection, and changepoint detection. The framework comes equipped with dedicated time series algorithms and scikit-learn compatible tools, enabling users to build, tune, and validate time series models efficiently. sktime also enhances interoperability by providing interfaces to related libraries such as scikit-learn, statsmodels, tsfresh, PyOD, and fbprophet, facilitating composite model building through features like pipelining, ensembling, tuning, and reduction.

text-generation-webui-extensions serves as a comprehensive directory for extensions designed to enhance the text-generation-webui. This resource enables users to significantly expand the capabilities and personalize the user interface of their text generation setups. The available extensions cover a wide range of functionalities, including Discord bots for both text and image generation, offering advanced features for various applications. The platform also encourages community contribution, allowing users to submit their own extensions to the growing list, fostering a collaborative environment for development and customization. This makes it a valuable hub for anyone looking to optimize and tailor their text generation experience.

TheWhisper is an open-source project dedicated to developing highly efficient speech-to-text and text-to-speech inference solutions, with a strong emphasis on self-hosting, cloud hosting, and on-device inference across various platforms. It provides optimized Whisper models with streaming inference support, offering flexible chunk sizes (10s, 15s, 20s, 30s) unlike the original 30s fixed size. The tool features high-performance inference engines for NVIDIA GPUs and CoreML engines for macOS/Apple Silicon, known for their low power consumption. It's ideal for real-time captioning, live meetings, voice interfaces, and edge deployments, and includes a local RestAPI with frontend examples and a demo Electron app for macOS.

verl, short for Volcano Engine Reinforcement Learning for LLMs, is an open-source RL training library designed for large language models. Initiated by ByteDance Seed team and maintained by the verl community, it provides a flexible, efficient, and production-ready framework for post-training. Key features include easy extension of diverse RL algorithms through its hybrid-controller programming model, seamless integration with existing LLM infrastructures like FSDP and Megatron-LM, and flexible device mapping for efficient resource utilization. verl is known for its state-of-the-art throughput and efficient actor model resharding with 3D-HybridEngine, significantly reducing memory redundancy and communication overhead. It supports various RL algorithms such as PPO, GRPO, and DAPO, and is compatible with popular Hugging Face and Modelscope Hub models.

WebUI is an open-source project built on Gradio, designed to enable users to run AI agents within their web browser. It leverages the browser-use foundation to make websites accessible for AI agents, offering a user-friendly interface for interaction. The tool boasts expanded support for numerous Large Language Models, including Google, OpenAI, Azure OpenAI, Anthropic, DeepSeek, and Ollama, with plans for further integration. A key differentiator is its custom browser support, allowing users to utilize their own browser instances, thereby bypassing re-authentication and enabling high-definition screen recording. Additionally, WebUI provides persistent browser sessions, allowing the browser window to remain open between AI tasks to maintain a complete history and state of AI interactions.

ZerePy is an open-source Python framework designed for deploying AI agents on the X platform, leveraging multiple large language models. Built from a modularized version of the Zerebro backend, ZerePy enables users to launch their own agents with similar core functionalities. It features a CLI interface for managing agents, a modular connection system, and blockchain integration for on-chain activities on Solana, Ethereum, and Monad. The framework supports various LLMs including OpenAI, Anthropic, Ollama, and XAI (Grok), and offers social platform integrations with Twitter/X and Farcaster. Users can customize agents with detailed configurations, including bios, traits, and examples, and integrate with the GOAT (Great Onchain Agent Toolkit) for advanced blockchain interactions.

Wuerstchen is an open-source framework designed for the efficient pretraining of text-to-image models. Unlike common approaches that use single-stage compression, Wuerstchen introduces an additional stage, resulting in a 42x compression factor while maintaining faithful image reconstruction. This multi-stage compression (Stage A, B, and C) allows the computationally expensive text-conditional part to be learned in a highly compressed latent space. The tool provides notebooks for both reconstruction (Stage B) and text-conditional generation (Stage C), and is fully integrated into the Hugging Face `diffusers` library, enabling easy use with Python. It also offers training scripts for users to train their own models, highlighting its speed and cost-effectiveness due to the smaller latent space (12x12).

Shelf is an operating system designed for agentic AI, enabling leading enterprises to design, build, and scale AI agents. It transforms business operations by accelerating workflows up to 60x and achieving up to 85% case resolution. The platform helps organizations reimagine how work gets done by optimizing and orchestrating AI agents, eliminating repetitive tasks, and providing full transparency into AI decisions. Shelf Agent OS transforms data into an AI-ready intelligence layer, unlocks intelligent experiences, and ensures reliable, compliant AI outcomes. It supports process automation and conversational AI across various industries.

AIRRIVED offers an enterprise agentic AI platform designed for building, fine-tuning, and orchestrating intelligent agents at scale. The platform aims to simplify the integration of AI into operational workflows, providing a robust solution for organizations looking to leverage agentic AI. It focuses on delivering capabilities that allow for the efficient management and deployment of AI agents across various systems, ensuring scalability and reducing complexity for enterprise-level applications. This makes it suitable for businesses seeking to unify and enhance their cybersecurity, IT, and general business operations through advanced AI.