Vercel Introduces Eve: An Agent Framework Built on Next.js’s File‑Based Model

File‑as‑Definition: A Familiar Front‑End Pattern

Vercel’s Eve framework applies the same “file equals definition” convention that made Next.js popular: placing a file in a specific directory automatically defines its role, eliminating the need for explicit routing tables or scaffolding.

Directory Layout of an Eve Agent

agent.ts : model configuration and Agent settings

instructions.md : system prompt that defines the Agent’s persona and rules

tools/ : each TypeScript file implements a tool; the filename hints at its purpose (e.g., tools/query-db.ts)

skills/ : Markdown files containing domain knowledge that teach the Agent how to handle specific scenarios

connections/ : integration files for external services such as MCP server or various APIs

channels/ : adapters for deployment targets like Slack, Discord, Teams—each represented by a single file

schedules/ : cron‑based scheduled tasks

subagents/ : nested Agent directories for task delegation and decomposition

Adding a new capability simply means dropping a .ts file into tools/; teaching new knowledge means writing a .md file in skills/. The mental model mirrors the experience of building a Next.js project.

Deployment and Front‑End Integration

An Eve Agent is just a regular Vercel project. It can be deployed with the standard Vercel command without any code changes. Locally it runs in a Docker sandbox; when pushed to Vercel it automatically switches to Vercel Sandbox. Sessions survive crashes and deployments, so a conversation that is interrupted by a new release continues on the previous deployment version.

For front‑end developers, the key convenience is the useEveAgent() hook. This hook queries the Agent’s HTTP API, and it works out of the box with frameworks such as Next.js, Vite, and TanStack, allowing a React component to attach an Agent to a chat UI with minimal code.

Real‑World Production Use at Vercel

Vercel is already running more than 100 production Agents on Eve, including:

d0 : a data‑analysis Agent that answers over 30,000 Slack questions per month

Vertex : a customer‑service Agent that resolves 92% of tickets automatically

Lead Agent : an SDR workflow Agent that costs about $5,000 per year, delivering roughly 32× ROI

Athena : a revenue‑operations tool built by non‑engineers

Vercel reports that roughly 29% of its deployments are Agent‑triggered today, and the team expects that share to approach 50% soon.

Critical Perspective

While Eve’s file‑system‑first approach solves the “reinventing the wheel” problem that plagued early Agent development, there are caveats. The framework is still in public preview with no announced pricing, so production‑grade cost estimates are uncertain. Moreover, Eve is tightly coupled to the Vercel platform—Sandbox, Gateway, and Workflow features run locally but require the full Vercel stack for a complete production experience, which may be a benefit for some teams and a lock‑in concern for others.

Ultimately, the value of a framework lies in the adoption of its conventions. Next.js succeeded by establishing a strong, simple convention; Eve aims to replicate that success in the Agent space. Front‑end developers should watch its evolution closely.