TL;DR / Key Takeaways
- Vercel just open-sourced Eve, a radical new framework for building production AI agents.
- Its 'file-system first' approach simplifies development while providing the reliability needed for millions of users.
The Folder is the Agent
Vercel's new open-source AI agent framework, eve, introduces a file-system first paradigm that fundamentally changes the way we build production agents. Traditionally, defining an entire AI agent by the contents of a single folder was reserved for personal assistants or "second brains." Scaling such a setup to production environments, serving thousands or millions of users, demanded complex, bespoke architectures.
eve bridges this gap, bringing the flexibility and ease of file-system-based development to robust production deployments. It provides a standardized, composable structure where an agent's entire definition resides within a single parent folder. This simplifies development, abstracting away much of the underlying plumbing.
Within this agent folder, discrete subfolders define core primitives, making everything modular and easy to manage. You configure: - Instructions: The system prompt and global rules. - Skills: Agent capabilities. - Tools: Sandboxed code execution environments. - Channels: Connections to platforms like Slack or Discord. - Sub-agents: Hierarchical agent structures. - Schedules: Autonomous execution timings. This architecture compiles seamlessly, eliminating manual imports or linking, providing a strong foundation for scalable, reliable AI systems.
Compilation Without the Complexity
Beyond the elegant folder-as-agent paradigm, eve's most impressive feature is its automatic compilation step. This completely changes the way we build production agents by eliminating the tedious, error-prone need for manual imports or explicit linking of components. Developers no longer write boilerplate code to connect their agent's various primitives; eve handles the complex plumbing under the hood.
When you run the `eve` command, the framework intelligently traverses the entire agent folder. It discovers all defined primitives—your instructions, agent definition, skills, tools, channels, connections, and sub-agents. From this traversal, eve dynamically creates a single manifest, a unified representation that hooks everything together, ready for robust deployment.
This workflow mirrors how certain coding agents, like Claude Code, handle skills: drop a skill file into a designated folder, and it becomes instantly available for invocation. But eve expands this concept dramatically. Instead of just skills, eve applies this 'just-works' magic to the entire agent architecture, from core logic to external integrations, making it viable for scalable production deployments and reliability.
From Your Laptop to Millions of Users
Vercel eve elevates agent development beyond hobbyist projects, leveraging Vercel's robust infrastructure to bridge the gap to enterprise applications. This completely changes the way we build production agents, enabling deployment to thousands or millions of users. You no longer need to worry about custom infrastructure or self-hosting; Vercel handles the entire hosting and scaling of your agent.
Production systems demand unwavering resilience. Vercel eve delivers with durable sessions, checkpointing every turn and tool call. This ensures workflows survive crashes and redeploys, allowing agents to resume naturally from any breaking point. Isolated sandboxing provides a secure environment for agent code execution, crucial for applications where agents might write or execute code autonomously.
Further enhancing reliability, eve incorporates human-in-the-loop functionality, pausing agents at risky steps for explicit approval. This mitigates the critical risk of autonomous agents running uncontrolled in production. For robust MLOps, eve introduces Evals as a deploy gate, allowing developers to define specific behavior tests directly within an agent's folder. Updates only push to production after achieving green check marks across these validations, guaranteeing consistent and reliable agent performance.
Build Your First Agent in Minutes
Building your first agent with Vercel eve shrinks development cycles to minutes, not days. The integrated Vercel plugin streamlines the entire process, from local iteration to global deployment, embodying a truly production-first developer experience.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
An agent's core begins with a simple `agent.ts` file, defining the foundational LLM. From there, you expand its capabilities by creating folders for skills, tools, connections, and even sub-agents. This file-system-first paradigm keeps everything modular and intuitively organized.
Defining new capabilities is as straightforward as dropping a Markdown or TypeScript file into the relevant folder. For instance, a `skill.md` file within a `/skills` directory instantly grants the agent a new ability, much like how Claude Code agents absorb new functions. This eliminates complex linking or import statements.
To test locally, simply run the `eve` command from your terminal. This command automatically compiles your agent, traversing all defined folders and hooking everything together without manual configuration. It's a seamless local feedback loop.
Once satisfied, deploying your agent to millions of users is a single-prompt operation through the Vercel plugin. This rapid deployment, coupled with Vercel's robust infrastructure, means you can move from an idea to a globally accessible, production-grade agent faster than ever before.
Frequently Asked Questions
What is Vercel Eve?
Vercel Eve is a new open-source, 'file-system first' framework for building production-grade AI agents. It standardizes agent development by treating the entire agent as a single, composable folder.
How is Eve different from other AI agent frameworks?
Unlike many frameworks that require complex code and manual linking of components, Eve automatically compiles everything in your agent's folder—skills, tools, prompts—into a single manifest. This dramatically simplifies the build process.
What makes Vercel Eve 'production-grade'?
Eve leverages Vercel's infrastructure to provide features essential for production, including durable sessions that survive crashes, isolated sandboxing for secure code execution, human-in-the-loop approvals, and Evals as a deploy gate.
Do I need a coding agent like Claude Code to use Eve?
While not strictly required, using a coding agent with the Vercel plugin is highly recommended. The plugin provides skills that let the agent understand Eve's structure, making it incredibly easy to build, configure, and deploy agents.