Skip to main content

Zavu + eve

eve is Vercel’s open-source agent framework. There are two ways to combine it with Zavu, depending on where you want the agent to run.

Option 1: deploy the eve project on Zavu

If your agent is instructions + tools, deploy it directly — it becomes a first-class Zavu agent answering on your senders:
See Deploy an eve project for what maps, what doesn’t, and the compatibility report.

Option 2: keep the agent on eve, call Zavu as a connection

If your agent relies on eve’s own runtime (sandbox, subagents, durable sessions), keep it there and give it Zavu’s messaging API as tools. eve’s OpenAPI connections turn our spec into one tool per operation:
The agent then gets tools like zavu__sendMessage. Recommendations:
  • Gate sends behind an approval so the model cannot message people unsupervised: add approval: once() (from eve/tools/approval) to the connection config if your review policy calls for it.
  • Scope the API key: create a key for this agent, and use a sub-account with a spending cap if the agent messages on behalf of a client.
  • sendMessage accepts phone numbers (E.164), email addresses, and chat IDs in the same to field, with channel: "auto" routing — one tool covers every channel.

Which option to pick

Both directions use the same account, senders, and message history — you can prototype with option 2 and migrate to option 1 (or the reverse) without changing anything else.