AI Agent Index — Coding Agents & Agent Platforms

The phenomenal open-source AI Agent project of 2026, written by Austrian developer Peter Steinberger (founder of PSPDFKit) over a single weekend. It racked up 180K Stars on GitHub in 30 days, ultimately surpassing 315K Stars to become one of the fastest-growing open-source projects in history (overtaking React's 13-year accumulation). Core positioning: 'an AI that actually gets work done' — local deployment, data stays yours, 20+ messaging platforms, YAML custom skills, 13000+ community plugins. Called 'the iPhone moment of AI Agents' by Andrej Karpathy; Tsinghua University published a dedicated research report

The hottest open-source project in workflow automation for 2026 (190K+ GitHub Stars), evolved from a 'no-code workflow tool' into an 'AI-native orchestration platform'. 400+ official integrations, native LangChain integration, MCP protocol support, AI Agent nodes. Visual node-wire editor with code-level extensibility. Self-hosting is completely free, data fully under your control. Increasingly used by enterprises to build complex AI automation workflows — the 'orchestration hub' connecting AI capabilities with business systems

The open-source LLM app development platform with the most GitHub Stars worldwide (passed 142K Stars in May 2026), offering full-stack capabilities from Agent building to AI Workflow orchestration, RAG retrieval augmentation, model management and observability analytics. Visual canvas drag-and-drop wiring lets you move from prototype to production without writing glue code. One-click Docker private deployment, data fully localized. No forced model lock-in; cost = platform fee + model Token fee (paid directly to vendors, Dify takes no cut)

ByteDance's one-stop AI Agent development & collaboration platform; v3.0 released June 1, 2026. Core upgrades are 'AI team collaboration' and 'project space': multiple AI Agents work together within a solo or group setup, synced across phone/PC/web. Zero-code drag-and-drop bot building, create an Agent in 30 seconds, the most complete Chinese ecosystem (Doubao/Jimeng/Jianying integrated), generous free quota. Positioned as 'say goodbye to tech dependency', cutting dev cycles from months to minutes

The framework with the richest AI Agent development ecosystem (GitHub 124K+ Stars), entering 2.x in 2026. LangChain provides chained calls / RAG / tool integration fundamentals; LangGraph 1.0 provides graph-based state-machine orchestration (supporting cycles / conditional branching / state recovery / human-in-the-loop nodes). LangSmith offers full-link observability. The ecosystem covers 1000+ integrations and underlies most Agent frameworks. Positioned as 'the React of AI apps' — not the end product, but the infrastructure for building everything

The world's first general-purpose AI agent, released March 2025 by the China-based team Monica.im (Butterfly Effect), igniting the global tech scene. Core breakthrough: multi-agent architecture + autonomous task execution, upgrading from 'passive tool' to 'proactive collaborator', with a full think→plan→execute closed loop, able to operate computers / analyze data / automate across platforms. Acquired by Meta for $2B in late 2025; Tencent redeemed it for ~¥13.6B in July 2026. Fully opened registration in May 2026, 300 free credits daily. It was acquired by Meta for over $2B in late 2025 and returned to independent operation on Sep 1, 2026; on Sep 28 it shipped 2.0 with the in-house Cascade agent framework (which the company says cuts token use 23.2%, task time 28.2% and run cost 32% in its single configuration), an upgraded Manus Studio, Cloud Computer and Automations, plus the Cue personal assistant.

Alibaba Cloud's one-stop large-model service platform, forming three subscription systems in 2026: Token Plan / Coding Plan / Agent Plan. The Token Plan Team edition uses unified Credits metering, integrating multi-vendor models such as Qwen/DeepSeek/Kimi/MiniMax/GLM, compatible with OpenAI/Anthropic standard interfaces, and adapted to mainstream AI tools like OpenClaw/Hermes Agent. Agent development supports visual workflow orchestration, RAG knowledge base and MCP protocol. Enterprise capabilities: multi-tenant isolation / SSO / data not used for training / no peak queuing

Meta's personal AI agent launched September 8, 2026, built on Muse Spark, Meta Superintelligence Labs' multimodal reasoning model. It breaks goals into subtasks, opens pages, fills forms and calls external apps to handle email, bookings, payments and even desktop control. It first rolled out in the US and Canada on iOS, Android, web and WhatsApp, and is woven into Messenger, Instagram, Facebook and AI glasses; 1.8M US/Canada iOS downloads in 12 days beat ChatGPT and topped the US App Store. Meta later open-sourced Muse Gadgets and shipped a Muse Home Link device for smart-home access.

Zhipu AI's general-purpose agent line, billed as the world's first phone agent: it drives a cloud-hosted virtual phone to order food, shop and run multi-step phone tasks, and generates PPTs, video, podcasts and web pages from one prompt. It reached version 2.0 in 2026, positioned against GPT-5 Agent Mode. The official entry autoglm.zhipuai.cn now also runs as a report assistant doing deep search and structured report generation.

Google Cloud's enterprise agent, announced October 8, 2026 at Gemini at Work 2026 and positioned as the universal agent for work: you set a goal instead of step-by-step instructions, and it handles knowledge work, content creation and coding while connecting to enterprise tools and running in the cloud for hours or even days with memory and context carried across devices. It launched with support for partner models including Claude, and is the core entry point of the Gemini Enterprise Agent Platform.

Microsoft's cloud-native AI Agent building platform, deeply bound to the Microsoft 365/Azure/Dynamics ecosystem. Low-code visual designer supporting natural-language Agent creation, connecting enterprise data sources (SharePoint/Teams/Outlook/CRM), and publishing to Teams/websites/Slack. In 2026 it deeply integrated with Microsoft Agent Framework 1.0 (the evolution of AutoGen/AG2), supporting multi-Agent orchestration. Enterprise-grade security & compliance (Azure AD / data residency / audit logging) make it the top choice for Microsoft 365 enterprise users building internal Agents

The most mature framework for multi-Agent collaboration (GitHub 45K+ Stars), with the core philosophy 'Don't build one agent, build a crew'. Designed around 'virtual teams': assign each Agent a Role, Goal and Tools, and let the platform automatically orchestrate collaboration. Supports both Crews (autonomous intelligence) and Flows (precise orchestration) modes, with an 82% task success rate. v1.14 released in 2026; the enterprise edition supports private deployment / SSO / audit. Ideal for fixed processes with clear division of labor: content creation / market research / report generation

An open-source AI knowledge-base Q&A system under the Apache 2.0 license (GitHub 27K+ Stars), focused on RAG retrieval-augmented generation. TypeScript + Next.js architecture, supporting MongoDB/PostgreSQL + Docker deployment. Core capabilities: out-of-the-box data preprocessing, visual Flow workflow orchestration, multi-model invocation, permission management. In 2026 it partnered with Sangfor to launch a commercial edition (Sangfor Agent Builder), supporting private deployment + self-learning engine. Widely used by developers worldwide for production-grade AI knowledge-base apps

Microsoft Research's open-source multi-Agent conversation framework (GitHub 53K+ Stars), officially renamed to AG2 and fully open-sourced in March 2025. Core trait: 'no forced workflow, fully customizable state' — it only provides Agent registration / message routing / conversation-loop fundamentals, leaving everything else to the developer. Multiple Agents reason and self-optimize through intelligent conversation, with a built-in safe code execution environment. In 2026 Microsoft Agent Framework 1.0 was released, deeply integrated with Azure and pushed into the enterprise mainstream

The most creative design among multi-Agent collaboration frameworks — it simulates a real software development process. Input a one-sentence requirement and it automatically assigns roles like product manager / architect / engineer / QA, collaborating via SOP (standard operating procedure) to output a complete software project (PRD → design → code → test). In 2026 it expanded to general multi-Agent tasks (data analysis / content creation / research), reaching GitHub 45K+ Stars. Core philosophy 'Code = SOP(Team)' — driving AI team collaboration the way human organizations work