Defining Two Emerging AI Agent Frameworks
ArrowJS and eve are open source agent tools that represent two competing AI agent frameworks: one rooted in reactive UI patterns and the other in directory-based agent architecture patterns that organize models, prompts, and tools as files. ArrowJS presents itself as a tiny reactive UI library recast as a framework for the agentic era, where both humans and large language models can understand the code through standard web primitives. Vercel’s eve treats each AI agent as a directory, compiling files into a durable workflow that can pause, resume, and run in production. Together, ArrowJS vs eve highlight a split between UI-centric and agent-centric design, giving teams a choice between front-end reactivity and structured workflow directories when they plan their next generation of AI-driven applications.
ArrowJS: Reactive UI Library for the Agentic Era
ArrowJS is a dependency-free reactive UI library created by Justin Schroeder that recently reached a stable 1.0 after three years as an experimental project. It focuses on core web technologies: JavaScript modules, tagged template literals, and the DOM, with no JSX, compiler, or required build step and a runtime under 5kb. Its entire public surface is three functions—reactive, html, and component—making it easy for both developers and coding agents to read and modify. According to InfoQ, ArrowJS benchmarks on par with Vue 3 while keeping the documentation compact enough to fit in less than 5 percent of a 200k token context window. The framework also offers an optional WASM sandbox via @arrow-js/sandbox to run untrusted agent-generated UI logic safely, plus layered packages for async components, server rendering, and hydration that support more complex AI agent frameworks without leaving the browser-centric model.
Eve: Agents as Directories and Durable Workflows
Eve is Vercel’s open source framework that treats an agent as a directory, echoing how Next.js defines a web app. Inside a single folder, one file selects the model, another defines the system prompt in Markdown, and each tool is a TypeScript file whose filename becomes the tool name with no extra registration. Eve compiles this directory into a running agent where every conversation is a durable workflow built on Vercel’s Workflow SDK, checkpointing each step so sessions can pause, survive crashes, and resume. The framework integrates with MCP servers and skill.md files, supports subagents, and provides channels for Slack, Discord, Microsoft Teams, Telegram, Twilio, GitHub, and Linear. Vercel reports that agents now trigger around 29 percent of deployments on its platform and expects this share to reach half, underscoring the production focus behind eve’s agent architecture patterns and observability story.
UI-First vs Agent-First Design: Architectural Trade-offs
Comparing ArrowJS vs eve highlights two different mental models for AI agent frameworks. ArrowJS is UI-centric: it makes reactivity the core abstraction, letting agents and humans manipulate DOM-bound state using plain JavaScript and tiny helpers. That makes it a strong fit when the primary challenge is building responsive, AI-generated interfaces that must run safely in the browser, especially with its QuickJS-based WASM sandbox. Eve, by contrast, is agent-centric: it centers the directory as the unit of design, where prompts, models, tools, and workflows live side by side as files. This layout suits teams whose main concerns are durable execution, tool orchestration, and cross-channel communication. Developers deciding between open source agent tools can ask whether their main complexity lies in UI rendering or long-running workflows and cross-system integrations, then pick the framework whose architecture matches that concern.
