Vercel has released Eve, a new open-source framework that reimagines how developers build and deploy AI agents by treating each agent as a directory of files, bundling all the infrastructure needed to run it in production. The framework, unveiled at Vercel's Ship conference in London on June 17, represents a significant shift in how agentic AI systems are structured and deployed, moving away from complex configurations toward a more intuitive, file-based approach.

How Does Eve Simplify Agent Development?

Eve works by organizing an agent's components into a single directory, much like how Next.js structures web applications. Each element of the agent lives in its own file, and the framework automatically compiles these files into a running agent without requiring separate registration or complex setup steps. This approach makes agent development more accessible to developers familiar with modern web frameworks.

• Model Configuration: A single file specifies which language model the agent runs on, with Vercel's AI Gateway handling automatic fallbacks between different providers if one becomes unavailable.
• System Prompt: The agent's instructions are written in Markdown format, making them easy to read and modify without learning specialized syntax.
• Tool Definition: Individual TypeScript files define each tool the agent can use, with the filename automatically becoming the tool's name, eliminating the need for manual registration.
• Skill Files and Integrations: Eve supports skill.md files and MCP (Model Context Protocol) servers to connect agents to external tools and services.

What Makes Eve's Execution Model Different?

Every conversation in Eve runs as a durable workflow, built on Vercel's open-source Workflow SDK, which checkpoints each step of the agent's execution. This means if an agent crashes or the system goes down mid-task, the session can pause and resume exactly where it left off, without losing progress or requiring the user to restart the conversation. This durability is critical for production deployments where reliability matters.

The framework also includes built-in security features. Each agent runs in its own sandbox to isolate any code it writes, preventing it from affecting other applications. Additionally, every tool can be configured to require human approval before execution, giving teams control over what actions agents can take autonomously.

How Do Agents Connect to the Outside World?

Eve agents can hand off work to subagents, connect to external services through MCP servers or OpenAPI documents, and reach users through multiple communication channels. The framework includes built-in integrations for Slack, Discord, Microsoft Teams, Telegram, Twilio, GitHub, and Linear, allowing agents to interact with teams wherever they already work.

For visibility and debugging, every agent run produces an OpenTelemetry trace that appears in Vercel's observability dashboard. Teams can monitor what their agents are doing and export detailed data to specialized services like Datadog and Honeycomb for deeper analysis.

What's the Competitive Landscape for Agent Frameworks?

Eve enters a rapidly crowded market of agent frameworks. Its closest TypeScript-native competitor is Mastra, a Y Combinator-backed framework that reached version 1.0 in January and is designed to run on any platform, unlike Eve's current Vercel-only deployment model. LangChain's LangGraph remains the most established agent framework overall, though it prioritizes Python and centers on the same durable execution approach that Eve offers. Inngest's AgentKit is another TypeScript option with built-in durability.

Cloud providers are also entering the space from the infrastructure side. Cloudflare builds agents on its Workers platform and Durable Objects, while Amazon's Bedrock AgentCore, Google's Vertex AI Agent Engine, and Microsoft's Agent Framework offer managed runtimes that can run agents from any framework. OpenAI's AgentKit, released last year, ties its tooling specifically to OpenAI's own models.

Why Is Vercel Investing Heavily in Agents?

Vercel's commitment to agent development is backed by real internal usage. The company runs more than 100 agents internally on Eve, including a data-analysis agent that employees query in Slack tens of thousands of times per month and a routing agent that directs questions to whichever agent can answer them. This internal deployment demonstrates the framework's maturity and practical value.

The growth of agent workloads on Vercel's platform is striking. Agents now trigger approximately 29 percent of all deployments on the platform, up from less than 3 percent a year ago, and the company expects that share to reach 50 percent. This trajectory suggests that agentic AI is becoming a core workload for the platform, not a niche use case.

Eve is available now in public preview and is licensed under Apache 2.0 on GitHub, making it freely available for developers to use and contribute to. Developers can start an agent locally with a single command and talk to it through a terminal interface, then deploy it using the same vercel deploy command used for any other project. Notably, if a new version of an agent ships while a session is mid-task, that session finishes on the version it started on, ensuring consistency.

Vercel has indicated that support for other platforms is coming, though for now Eve runs exclusively on Vercel's infrastructure. This limitation may change as the framework matures and the company expands its deployment options.