Vercel Introduces Eve, an Open-Source Framework for Building AI Agents

Vercel has released Eve, an open-source framework for building, deploying, and operating AI agents in production. The framework provides a filesystem-based project structure that organizes an agent into directories for instructions, tools, skills, subagents, communication channels, and scheduled tasks, allowing developers to define an agent's behavior without manually implementing much of the supporting infrastructure.

Eve is designed around a filesystem-first architecture, where each capability is represented as a file. Developers configure an agent through a single configuration file, define its behavior using Markdown instructions, add tools as TypeScript files, and organize reusable knowledge as skills. During build time, the framework automatically discovers these components and exposes them to the agent without additional registration code.

The framework includes several production-oriented capabilities, including durable execution, sandboxed code execution, human approval workflows, subagents, tracing, and evaluation tools. Each conversation is stored as a durable workflow that can pause, survive failures or deployments, and resume from the last completed step. Agent-generated code executes inside isolated sandboxes that can run locally using Docker or other adapters, or in production using Vercel Sandbox.

Eve also provides built-in support for connecting agents to external services through Model Context Protocol servers or APIs described with OpenAPI specifications. At launch, integrations include services such as Slack, GitHub, Snowflake, Salesforce, Notion, and Linear, while developers can add additional providers through custom adapters. The same agent can be deployed across multiple communication channels, including Slack, Discord, Microsoft Teams, Telegram, GitHub, and HTTP APIs, without changing the core implementation.

The framework includes observability features based on OpenTelemetry. Every model call, tool invocation, and sandbox command is recorded as a trace that can be exported to existing monitoring platforms or viewed through Vercel's observability interface. Developers can also define evaluation suites that verify agent behavior locally or as part of continuous integration pipelines before deployment.

Another capability is support for subagents. Developers can define specialized agents in separate directories with their own instructions, tools, and execution environments. Parent agents can delegate work to these subagents, which execute tasks independently before returning results to the main agent. Eve also supports scheduled execution through cron-based jobs, enabling agents to perform recurring tasks automatically.

Vercel said the framework is already used internally to operate more than one hundred production agents supporting functions such as analytics, customer support, sales operations, and content review. The company has released Eve in public preview together with a command-line tool that scaffolds new agent projects and deploys them using standard Vercel workflows.

Early community reactions have been largely positive. Developers highlighted Eve's filesystem-based design and its inclusion of production features such as durable execution, tracing, approvals, and sandboxing.

AI Software Engineer, Michael Yagudaev shared:

Looks simple, and follows similar ideas to Claude Code and other agents. Like Pi-mono but also for multi-tenant, How do you deploy it? I’m guessing there are Vercel primitives for it?

Others compared it to frameworks noting that Eve attempts to consolidate infrastructure that many teams currently assemble from multiple libraries. Some commenters also expressed interest in evaluating how portable the framework will be outside the Vercel ecosystem as adoption grows.

Software developer Vikranth Kanumuru commented:

I don’t get what is this changing from existing frameworks. Are you even removing deterministic orchestration to be more unpredictable? And hand it off to a subagent?

Eve enters a growing ecosystem of agent frameworks that includes LangGraph, CrewAI, AutoGen, and Strands Agents SDK. While these frameworks also support multi-agent orchestration and tool integration, Eve distinguishes itself by combining durable execution, sandboxed code execution, built-in evaluations, approvals, scheduling, and deployment into a single framework organized around the filesystem.