eve: AI agents without the plumbing
The same agent, built twice: once without eve, once with it. Before a support agent answers a single customer, you need a Slack app with tokens and webhooks, auth for your billing providers, somewhere safe to run code, a cron job, crash recovery, and a human in the loop before any money moves. That is plumbing, not the agent. With eve, Vercel's open source agent framework, the scaffolding is already there. The agent itself is just a directory: one instruction.md file and five folders. Vercel runs hundreds of eve agents in production today. WHAT'S COVERED ▲ Channels: Slack, Discord, Teams, Telegram, or SMS, one file each ▲ Connections: MCP or OpenAPI auth, the model never sees a credential ▲ Tools and Skills: TypeScript in a Vercel Sandbox, policies as markdown ▲ Approval gates: refunds park in the thread until a human clicks approve ▲ Durable workflows: every turn checkpointed, crashes pick up where they left off CHAPTERS 0:00 The same agent, built twice 0:19 The plumbing every agent needs 0:42 With eve, an agent is a directory 0:54 The five folders in action 1:44 Gated approvals and durable workflows 2:23 Own the agent, not the plumbing 2:40 Hundreds in production, open source, one command away LINKS Eve → https://eve.dev Ship 2026 recap → https://vercel.com/blog/vercel-ship-2026-recap Start building → https://vercel.com FOLLOW VERCEL X → https://x.com/vercel LinkedIn → https://www.linkedin.com/company/vercel GitHub → https://github.com/vercel #Vercel #Eve #AIAgents #AgentFramework #OpenSource