Vercel has officially launched the eve agent framework, which is an open-source tool specifically designed for building, running, and scaling artificial intelligence agents. By utilizing this framework, developers are able to define agent behavior through a directory-based file structure, eliminating the need to manually assemble production infrastructure.
Core Features of the eve agent framework
The framework includes several built-in production features, such as durable execution, sandboxed compute, and human-in-the-loop approvals. Within this system, every conversation functions as a durable workflow. This design allows active sessions to pause, survive system deployments, and resume later using the open-source Workflow SDK. To ensure security, any agent-generated code runs in an isolated sandbox environment that is kept entirely separate from the main application runtime. Developers can utilize Vercel Sandbox, Docker, or microsandbox to run these isolated environments.
Connectivity and Channel Support
To simplify connectivity, the eve agent framework makes it easier to connect to external services using either Model Context Protocol (MCP) servers or OpenAPI-compatible APIs. At the time of its launch, the framework supports direct integrations with major platforms, including Slack, GitHub, Snowflake, Salesforce, Notion, and Linear. Furthermore, the same AI agent can serve multiple communication channels simultaneously. These channels include Discord, Teams, Telegram, and Twilio, and they are supported through the use of small adapter files.
Tracing and Evaluation Capabilities
For monitoring and debugging, the framework automatically generates OpenTelemetry-compliant traces for every single run. These traces map out model calls, tool executions, and sandbox commands in detail. Developers have the option to export these traces to external monitoring services, such as Datadog, Honeycomb, or Jaeger. Additionally, the system includes a dedicated testing suite called evals. This suite allows development teams to write scored checks to test agent behavior either locally on their machines or within continuous integration pipelines.
Internal Deployment and Market Growth
Vercel reported that it is already using the eve agent framework extensively for its own operations. The company currently runs more than 100 agents internally to manage its daily tasks, which includes a specific data analyst agent that processes over 30,000 questions every month. According to Vercel, agent-related deployments on its platform have experienced rapid growth, increasing from less than 3% last year to approximately 29% today. Looking ahead, the company expects that AI agents will eventually drive half of all deployments on its platform.
Source: Vercel
