Executive Summary
Vercel has released "eve" into public preview, an open-source, filesystem-first framework designed for building, running, and scaling AI agents. The framework simplifies agent creation by defining them as a directory of files for their model, instructions, and tools. eve is built for production environments, with features like durable execution, sandboxed compute, and human-in-the-loop approvals included by default, and can be deployed directly on Vercel.
Key Takeaways
* Product: eve, an open-source framework for building durable AI agents.
* Structure: Agents are defined as a simple directory of files, specifying their model (`agent.ts`), instructions (`instructions.md`), tools, skills, and channels.
* Built-in Production Features: Includes durable execution, sandboxed compute, human-in-the-loop approvals, subagents, and evaluation tools.
* Developer Experience: A new agent can be scaffolded and run locally with a single command (`npx eve@latest init`).
* Deployment: As a standard Vercel project, an eve agent can be deployed to production using `vercel deploy`.
* Compatibility: Works with any language model and can be integrated with channels like Slack, Discord, and GitHub.
* Availability: Now available in public preview.
Strategic Importance
By open-sourcing the framework it uses for its own agents, Vercel is positioning itself as a foundational platform for AI application development, aiming to standardize the workflow for building and deploying production-ready agents within its ecosystem.