Featured Agents
Discover the general AI agents driving the industry forward. We highlight notable open-source frameworks and commercial platforms, breaking down what they do and how they are deployed. Explore this curated list to find the digital workers and developer tools that best fit your workflows.
Commercial
15Manus is a fully managed cloud-based general-purpose AI Agent developed by Manus AI and later acquired by Meta. Its tagline — "Hands On AI" — reflects its core mission: not just answering questions, but actually executing tasks, automating workflows, and extending human reach. Manus runs in a cloud sandbox environment, capable of autonomously completing multi-step tasks such as research, writing, website building, and data analysis.
OpenAI Codex is an AI coding agent platform launched by OpenAI in April 2025. It wraps frontier models with file access, shell execution, sandboxes, approval flows, and code review into a complete developer workflow. Codex runs across desktop (macOS/Windows), CLI (open-source, Apache 2.0), web, and IDE extensions (VS Code, Cursor, Windsurf), and supports multi-agent parallel execution — developers can run several coding agents simultaneously on independent threads. As of March 2026, it has surpassed 2 million weekly active users.
Claude Cowork is Anthropic's general-purpose AI agent that brings Claude Code's agentic capabilities to the desktop for non-coding knowledge work. Launched in January 2026, Cowork can autonomously work with your local files — reading, editing, creating, and organizing documents — while you focus on other tasks. It handles browser automation via a Chrome extension, integrates with third-party services (Asana, Notion, Zendesk, Microsoft Teams), and can spawn sub-agents to tackle complex multi-step projects from different perspectives simultaneously. Built on the same Claude Agent SDK that powers Claude Code, Cowork runs in an isolated sandbox using Apple's Virtualization Framework for security.
Gemini Spark is Google's first persistent, 24/7 autonomous AI agent, announced at Google I/O 2026. Unlike traditional chatbots, Spark runs on dedicated virtual machines in Google Cloud, continuing to execute tasks even when devices are off. It combines three core capabilities — Tasks (autonomous multi-step execution), Skills (learning personal style and preferences), and Schedules (time-based or conditional triggers) — and deeply integrates with Google Workspace (Gmail, Calendar, Docs, Drive, Sheets, Slides). Powered by Gemini 3.5 Flash and built on the Antigravity agentic platform, Spark supports up to 15 simultaneous tasks, third-party integrations (Canva, OpenTable, Instacart, with Uber and Spotify coming), and web browsing via Chrome.
Marvis is Tencent's operating system-level AI assistant, officially launched in May 2026. Unlike chatbot-style agents, Marvis ships with a pre-installed team of 6 AI agents — Master (task orchestration), File (semantic search, editing, format conversion), Computer (system O&M, hardware diagnostics), App (cross-application operations), Browser (web interaction), and Search (information aggregation) — that work in parallel to complete complex tasks. It deeply integrates with Windows, Mac, and Android, offering both Efficiency Mode (cloud-edge collaboration powered by Hunyuan and DeepSeek V4) and Privacy Mode (fully local via Qwen, works offline). Users get 10 million free tokens per day.
Genspark Super Agent is a hybrid-architecture AI Agent developed by Genspark. It employs a MoA (Mixture of Agents) architecture, combining multiple specialized agents working together to complete deep research, cross-application collaboration, and even real-world tasks like making phone calls.
Accio Work is Alibaba International's enterprise-grade AI agent for global e-commerce, launched in March 2026. Built on top of the Accio B2B AI search engine (Nov 2024), it orchestrates a team of specialized sub-agents — Procurement, Negotiation, Logistics, and Store Operations — to autonomously run an end-to-end cross-border business. From a single idea, Accio Work can conduct market research, design products, source suppliers, negotiate deals via email, manage logistics, and even build a complete Shopify store in 30 minutes. It connects to Gmail, Google Drive, Notion, Telegram, Discord, DingTalk, and WeChat, powered by Qwen2.5 and DeepSeek reasoning models. As of 2026, it serves over 10 million enterprise users worldwide.
TRAE SOLO is ByteDance's responsive AI coding agent and productivity workbench, launched as a standalone product in March 2026. Built on TRAE's IDE foundation (beta July 2025, official November 2025), SOLO flips the traditional developer workflow — AI leads execution while humans provide requirements and review results. It offers two built-in agents (SOLO Coder for complex projects with Plan mode and sub-agent orchestration, SOLO Builder for rapid end-to-end app generation) and two work modes: Code Mode for developers and Work Mode (MTC) for PMs, analysts, and operations. Key capabilities include context engineering across 100K code files, master-sub-agent architecture, context compression, multi-task parallel execution via cloud computing, DiffView for code changes, voice interaction, and three-device sync (desktop, web, mobile).
Perplexity Comet is a browser-assisted research Agent developed by Perplexity AI. It opens multiple tabs in the browser for deep search, automatically collects information, and generates comprehensive research reports with cited sources. Comet is designed as a researcher's assistant, specializing in multi-perspective, multi-source information retrieval and synthesis.
AGI-0 is a web-controlling Agent developed by AGI, Inc. (formerly MultiOn). Given a goal, AGI-0 autonomously completes web-based tasks such as e-commerce shopping, information lookup, and form filling. AGI-0 ranks #1 on multiple major computer control benchmarks and can be deployed on phones, laptops, tablets, and other smart devices.
Lindy.ai is a no-code AI virtual employee (AI Executive Assistant) focused on automating daily workflows including email, calendar, customer support, and market research. Lindy is designed as your "virtual executive assistant," accessible via iMessage, SMS, or a web interface. It aims to save you approximately 2 hours of office work each day.
Taskade is a team collaboration tool infused with AI capabilities, integrating project management, document collaboration, and AI Agent features into a unified workspace. Taskade embeds a multi-agent system that enables teams to leverage AI assistants for task management, meeting notes, web scraping, and more.
Relevance AI is an AI Workforce Platform that allows users to build general-purpose AI agents through a no-code interface, targeting sales, research, operations, and more. The platform enables users to create "AI team members," where each agent can be assigned specific roles, tools, and knowledge bases to execute end-to-end workflows autonomously.
Kimi Explore is a general-purpose research agent developed by Moonshot AI (月之暗面), featuring deep autonomous search and multi-turn research capabilities. Unlike ordinary AI chat assistants, Kimi Explore possesses proactive search, information synthesis, and multi-step reasoning abilities, enabling it to autonomously complete complex research tasks like a research assistant.
Coze is an AI Agent building and usage platform developed by ByteDance. It serves both as a consumer marketplace for AI agents (Bot Store) for everyday users and a comprehensive agent-building platform for developers and enterprises. Coze provides a complete toolchain including knowledge bases, workflows, plugin systems, and multi-model support.
Open Source
15OpenClaw is a highly popular open-source personal AI agent that runs locally on the user's desktop with over a hundred skills for executing various system tasks. OpenClaw's design philosophy is "The AI that actually does things" — it can clear your inbox, send emails, manage your calendar, check you in for flights, and handle many everyday operations.
Hermes Agent is an open-source general-purpose agent orchestration system created by Nous Research. Designed as "the agent that grows with you," it provides core capabilities including tool calling, sub-agent management, memory systems, and skill extensibility, enabling autonomous task execution across programming, research, writing, data analysis, and more.
AutoGPT is the originator and most iconic project among open-source autonomous AI agents. Developed by Significant Gravitas, AutoGPT's mission is to provide "accessible AI for everyone" — enabling AI to autonomously analyze, decompose, and execute complex tasks. Given a goal, it autonomously thinks about what to do next, executes actions, observes results, and iterates until the task is complete.
OpenHands (formerly OpenDevin) is an open-source autonomous software engineering agent focused on enabling AI to perform development tasks like a human software engineer. It operates in a sandboxed environment where it can write code, debug, use the command line, browse the web, and manipulate files. It is the best-known open-source alternative to the commercial product Devin.
DeerFlow is an open-source long-horizon SuperAgent Harness created by ByteDance, positioned as a general-purpose AI agent framework capable of handling almost any task. By orchestrating sub-agents, managing memory systems, providing sandboxed environments, and offering an extensible skill system, it can handle everything from research and reports to slides, web pages, images, and video generation.
Open Interpreter is an open-source, locally-run, code-driven AI agent that lets users control their computer using natural language. It executes Python, JavaScript, Shell, and other code locally to handle file operations, data analysis, web browsing, system administration, and various other tasks.
OpenManus is an open-source replica of Manus, built by the MetaGPT team. The prototype was launched within 3 hours and has been continuously iterating. Its motto is "No fortress, purely open ground" — emphasizing openness and no barriers. It supports web browsing, file operations, and code writing as general AI agent capabilities.
Goose is a local AI agent project incubated by the Agentic AI Foundation (AAIF) under the Linux Foundation. It is a general-purpose AI agent that runs on your machine — not just for code, but for research, writing, automation, data analysis, and any task you need to get done. Built in Rust with deep MCP integration.
CowAgent is an open-source super AI assistant (formerly chatgpt-on-wechat), created and maintained by zhayujie. It's not just a chatbot — it's a complete Agent Harness reference implementation that proactively plans tasks, controls your computer and external services, creates and runs skills, and grows alongside you through a personal knowledge base and long-term memory.
Nanobot is an open-source, ultra-lightweight agent runtime created by the HKU Data Science (HKUDS) team. It's designed for people who want to own their AI agent stack, providing a small, readable core plus practical pieces for real long-running agents: WebUI, chat channels, tools, memory, MCP, model routing, and deployment.
ZeroClaw is a high-performance, fully autonomous personal AI assistant infrastructure written in Rust. It's a single Rust binary you configure and run — becoming your personal AI assistant. ZeroClaw is a member of the OpenClaw ecosystem with a security-first architecture.
OpenHuman is a desktop universal agent built with Rust by the tinyhumansai team. It runs locally, supports 30+ model providers, has a 1-billion-token local memory tree, and integrates 118+ third-party services through cross-app tools. The core philosophy is "context in minutes, not weeks."
AgenticSeek is a 100% local alternative to Manus AI, created by developer Fosowl. This voice-enabled AI assistant autonomously browses the web, writes code, and plans tasks while keeping all data on your device — ensuring complete privacy and zero cloud dependency. It is tailored for local reasoning models.
QwenPaw (formerly CoPaw) is your personal AI assistant, built by the AgentScope team on the Qwen ecosystem. It's easy to install, deploy locally or in the cloud, connect across channels, and extend with skills. The philosophy is "Works for you, grows with you" — from simple conversation to complex multi-agent collaboration.
GenericAgent is a minimal, self-evolving autonomous agent framework. Its core is just ~3,300 lines of seed code. Through 9 atomic tools + a ~100-line Agent Loop, it grants any LLM system-level control over a local computer — covering browser, terminal, filesystem, keyboard/mouse input, screen vision, and mobile devices (ADB). The design philosophy is "don't preload skills, evolve them."