Introducing eve, an open-source agent framework
By Steven Van ·
Vercel's new open-source framework turns a directory of files into a production-ready agent, with durable execution and sandboxing built in.
Vercel has released eve, an open-source framework for building, running and scaling agents, now available in public preview. In eve, an agent is a directory of files: a model and an instructions file are enough to run one, and adding a tool, skill, channel or schedule is just a matter of adding a file, since eve wires each one in at build time without extra boilerplate.
Production concerns are built into the framework rather than left to the developer, including:
- Durable execution, so a session can pause, survive a crash or deploy, and resume where it stopped
- Sandboxed compute, so agent-generated code runs isolated from the application runtime
- Human-in-the-loop approvals for actions that need sign-off before running
- Subagents that a parent agent can delegate work to
- Evals for scoring agent behaviour
Running npx eve@latest init installs dependencies, scaffolds a project and starts a dev server, giving a working local agent in under a minute; a coding agent can also be prompted to set one up. Because an eve agent is an ordinary Vercel project, vercel deploy ships it to production unchanged. eve is the framework Vercel uses to build and run its own agents, and development happens in the open on GitHub. Read the announcement for more.
