Gemini CLI is Google's open‑source AI agent that integrates Gemini 2.5 Pro into the terminal. It supports a 1 million‑token context window, reasoning-and-act loops, Model Context Protocol, web grounding with Google Search, multimodal tasks (code, content, images, video), file manipulation, shell execution, and integration with Gemini Code Assist and MCP tools. It works on Windows, macOS, and Linux.

Gemini CLI

gemini-cli

An open‑source command‑line AI agent bringing Gemini 2.5 Pro into your terminal for coding, research, content and task automation.

Kimi‑Dev (72B) is an open‑source coding language model by Moonshot AI, fine‑tuned via reinforcement learning to autonomously fix bugs and write tests, achieving state‑of‑the‑art SWE‑bench performance.

Kimi‑Dev

kimi-dev

Open‑source coding LLM optimized for automated debugging and issue resolution tasks.

MiniMax‑M1 is an open‑weight, hybrid Mixture‑of‑Experts large language model offering natively a 1 million token context window. It uses lightning attention and efficient RL training (CISPO) to excel at long-context reasoning, tool-use, coding, and mathematics. Released under Apache 2.0, it supports structured function calling and is deployable via vLLM, Transformers, or API.

MiniMax‑M1

minimax-m1

Open‑source large‑scale reasoning model with 1 million token context and hybrid Mixture‑of‑Experts architecture.

Chroma is an open-source, AI-native vector database built for storing and querying embeddings alongside metadata. It supports full‑text and multimodal search, scalable retrieval, and powers retrieval‑augmented generation workflows. Available as both a local SDK and a private‑preview managed cloud, Chroma offers strong developer ergonomics and extensible integrations for building production‑grade LLM applications.

Chroma

chroma

An open‑source vector database and embeddings engine for building retrieval‑augmented AI applications.

Suna AI, developed by Kortix, is a fully open-source generalist AI agent designed to autonomously perform complex real-world tasks based on natural language instructions. It integrates capabilities such as web browsing, data extraction, document generation, and system-level operations, enabling users to delegate multifaceted workflows. With a modular architecture built using TypeScript and Python, Suna supports browser automation, command-line execution, and API integrations. Users can deploy Suna via a cloud interface or self-host it for enhanced control and customization.

Suna by Kortix AI

suna-ai

An open-source generalist AI agent that autonomously executes complex real-world tasks through natural language interaction.

Refact.ai is an open-source AI coding assistant designed to enhance developer productivity by providing intelligent code completions, refactoring tools, and context-aware chat within popular IDEs like JetBrains and Visual Studio Code. Utilizing Retrieval-Augmented Generation (RAG), it understands entire codebases to deliver accurate suggestions. Refact.ai supports over 25 programming languages and allows for self-hosting, ensuring data privacy. Its flexible pricing includes a free tier with 5,000 coins and a Pro plan offering additional features and coin allowances. Enterprise solutions include on-premise deployment and model fine-tuning tailored to organizational codebases.

Refact.ai

refact-ai

An open-source AI coding assistant offering context-aware code completions, refactoring, and in-IDE chat across major IDEs.

Agent Network Protocol (ANP) is an open-source communication protocol designed to facilitate secure, decentralized interactions among AI agents. It aims to become the 'HTTP of the Agentic Web Era' by providing a standardized framework that enables agents to connect, communicate, and collaborate efficiently. ANP's architecture comprises three layers: the Identity Layer for decentralized authentication using W3C DID standards, the Meta-Protocol Layer for dynamic protocol negotiation, and the Application Layer for semantic-based capability descriptions. This structure ensures interoperability, scalability, and secure communication across diverse AI systems.

Agent Network Protocol

agent-network-protocol

An open-source protocol enabling secure, decentralized communication among AI agents.

Hugging Face's Open Computer Agent is an open-source AI tool designed to perform web-based tasks by emulating human interactions within a virtual Linux desktop environment. Powered by vision-language models like Qwen2-VL-72B and frameworks such as smolagents and E2B Desktop, it can navigate websites, fill out forms, and retrieve information based on natural language prompts. Operating through a browser interface, the agent simulates mouse and keyboard actions to execute tasks. While still in its experimental phase, it showcases the potential of AI agents in automating routine digital activities.

Hugging Face Open Computer Agent

hugging-face-open-computer-agent

An open-source AI agent that automates web tasks by simulating user interactions in a virtual Linux environment.

Google's Agent Development Kit (ADK) is an open-source framework designed to simplify the creation, deployment, and management of AI agents. Optimized for Google's Gemini models yet model-agnostic, ADK supports various agent types, including LLM-powered agents, workflow controllers, and custom agents. It offers features like multi-agent orchestration, tool integration, session management, and built-in evaluation tools. ADK facilitates flexible deployment options, including local environments, Google Cloud's Vertex AI Agent Engine, and Cloud Run, making it suitable for a wide range of AI applications.

Google Agent Development Kit

google-agent-development-kit

A modular framework for building, deploying, and managing AI agents across diverse workflows.

Google's Agent2Agent (A2A) protocol is an open standard designed to facilitate seamless communication and collaboration between AI agents developed by different organizations using various frameworks. A2A provides a standardized method for agents to discover each other's capabilities, negotiate interaction modalities, and securely collaborate on tasks without exposing internal states or proprietary tools. Built on existing technologies like HTTP, JSON-RPC 2.0, and Server-Sent Events (SSE), A2A supports both synchronous and asynchronous interactions, enabling agents to handle complex, long-running tasks efficiently. The protocol emphasizes security, interoperability, and scalability, making it suitable for enterprise-grade applications.

Google A2A

google-a2a

An open protocol enabling AI agents to communicate and collaborate across diverse systems securely.

Crawl4AI is an open-source, LLM-friendly web crawler and scraper designed to streamline web data extraction for AI agents and data pipelines. It offers features such as asynchronous crawling, structured data output in formats like JSON and Markdown, and integration capabilities with large language models. Crawl4AI supports advanced extraction strategies, including LLM-based methods, and provides tools for handling dynamic content, making it a versatile solution for developers seeking efficient and customizable web scraping solutions.

Crawl4AI

crawl4ai

An open-source, LLM-friendly web crawler and scraper optimized for AI agents and data pipelines.

Dcup is an open-source, self-hostable Retrieval-Augmented Generation (RAG) platform designed to help developers and teams build scalable, AI-powered retrieval pipelines. It enables seamless integration with data sources like Google Drive, Dropbox, and AWS, and supports advanced search capabilities including LLM re-ranking, summary indexing, and hybrid search using OpenAI embeddings and Qdrant vector storage. With its intuitive Retrieval API and modular architecture, Dcup simplifies the process of ingesting, indexing, and retrieving data, making it ideal for AI applications requiring efficient and accurate information retrieval.

Dcup

dcup

An open-source RAG-as-a-Service platform for building scalable, AI-powered retrieval pipelines.

Explore open-source AI Agents