Vercel's Eve framework treats AI agents as directories

Key Takeaways

• Eve uses a filesystem-first approach where agents are defined by directory structure, not configuration files
• The framework supports durable tasks up to 24 hours via Vercel Workflows
• Eve is fully open-source under the MIT license, following Vercel's Next.js playbook

Vercel has released Eve, an open-source framework that simplifies AI agent development by treating agents as directories on disk. Instead of wiring together configuration files and orchestration logic, developers define an agent's instructions, tools, and skills through file placement. The approach borrows directly from Next.js's file-based routing, a pattern that helped make React accessible to millions of developers.

The release comes as the AI agent space grows increasingly crowded. AWS just announced AgentCore at its NYC summit, LangChain continues to evolve, and startups are shipping agent frameworks weekly. Vercel's bet is that developer experience wins, and Eve is designed to feel familiar to anyone who has built a Next.js app.

How does the filesystem-first approach work?

Eve's core idea is convention over configuration. An agent lives in a directory. Inside that directory, you place files that define its behavior. An instructions file tells the agent what it does. A tools directory contains the functions it can call. Skills are reusable capabilities that can be shared across agents.

This structure means a new team member can open a project and understand what each agent does without reading documentation or tracing through initialization code. The filesystem is the source of truth.

Compare this to most agent frameworks, which require developers to instantiate agents in code, pass configuration objects, and manually wire up tool registrations. Eve eliminates that boilerplate. You create a folder, drop in the right files, and the framework handles the rest.

What makes Eve agents durable?

AI agents often need to run long tasks. They might wait for human approval, poll external APIs, or orchestrate multi-step workflows that take hours. Traditional serverless functions time out after seconds or minutes. Eve solves this through integration with Vercel Workflows.

Tasks can run for up to 24 hours, pausing and resuming as needed. The state persists across function invocations. This durability is essential for production agent deployments where failures mid-task would mean lost work or inconsistent data.

Why is Vercel open-sourcing this?

Eve ships under the MIT license, fully open-source. The strategy mirrors what Vercel did with Next.js: give away the framework, build a community around it, and monetize through hosting and infrastructure. Developers who build with Eve will likely deploy on Vercel.

This playbook has worked before. Next.js now powers a significant portion of React deployments, and Vercel's revenue has grown accordingly. By establishing Eve as a standard for agent development, Vercel positions itself as the default deployment target for the next wave of AI applications.

How does Eve compare to other agent frameworks?

The agent framework landscape is crowded. LangChain offers flexibility and a massive ecosystem. CrewAI focuses on multi-agent collaboration. AutoGen from Microsoft targets research and experimentation. AWS AgentCore, announced just this week, provides enterprise-grade tooling with deep AWS integration.

Eve's differentiator is simplicity. It does not try to be everything. It provides a clear structure, sensible defaults, and tight integration with Vercel's deployment infrastructure. For teams already using Next.js and Vercel, Eve requires almost no learning curve.

Early reactions on Hacker News have been positive. Developers are comparing Eve to the early days of serverless, when conventions around function placement and deployment transformed backend development. Some see potential for Eve's directory-based approach to become a de facto standard.

What are the limitations?

Eve is new. The framework lacks the ecosystem depth of LangChain and the enterprise integrations of AWS AgentCore. Teams needing advanced features like built-in memory systems, complex agent-to-agent communication, or specialized connectors may find Eve too minimal for their needs.

The tight coupling with Vercel's infrastructure is both a strength and a constraint. While you can run Eve elsewhere, the durability features depend on Vercel Workflows. Organizations committed to other cloud providers may prefer frameworks that are cloud-agnostic.

Who should use Eve?

Teams building their first production agent and already comfortable with Vercel's ecosystem will find Eve the fastest path to deployment. Startups that need to ship quickly without deep infrastructure expertise will appreciate the conventions. Projects where agent logic is straightforward, think customer support bots, content generation pipelines, or internal automation, fit well.

Enterprise teams with complex requirements, strict multi-cloud policies, or existing investments in other frameworks should evaluate carefully. Eve's simplicity comes from making decisions for you. If those decisions do not match your constraints, the framework becomes a liability.

Frequently Asked Questions

Is Eve only for Vercel deployments?

Eve is open-source and can run elsewhere, but its durable task features rely on Vercel Workflows. Full functionality requires Vercel's infrastructure.

What programming language does Eve use?

Eve is built for JavaScript and TypeScript developers, consistent with Vercel's ecosystem.

How does Eve handle agent memory and state?

Durable state is managed through Vercel Workflows, which persist task progress across function invocations for up to 24 hours.

Can Eve agents call external APIs?

Yes. Tools defined in the tools directory can call any external API, database, or service the agent needs to complete its tasks.

Source: The New Stack / Frederic Lardinois