Novu Agent Inbox separates human replies from optional agent execution

[Novu’s November 2 announcement](https://www.novu.co/changelog/) gives an existing agent its own email address, Telegram bot and Slack app through CLI or MCP. Agent Inbox can ask people for a decision and [report approval, denial, waiting or expiry](https://www.novu.co/changelog/agent‑inbox‑give‑your‑ai‑agent‑a‑way‑to‑reach‑you/) back to the calling workflow.

[Coding agents normally check messages only while running](https://www.novu.co/changelog/agent‑inbox‑give‑your‑ai‑agent‑a‑way‑to‑reach‑you/). Novu’s optional Claude Code listener can instead start work in a chosen project when the owner or a selected contact writes. Listening must be approved first; this is not automatic execution for every connected agent.

A communication address and an execution trigger are separate configuration choices. Novu describes [initial email reception as a free demo until the inbox is claimed](https://www.novu.co/changelog/agent‑inbox‑give‑your‑ai‑agent‑a‑way‑to‑reach‑you/), which does not establish unlimited free service. Engineers pair programming, December 2012; archival context, not Novu staff or the Agent Inbox interface. Photo: Vajrapani666 / [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:Pair_Programming_3.jpg), [CC BY‑SA 3.0](https://creativecommons.org/licenses/by‑sa/3.0). Display crops retain the same licence.