Launching today
Agent Inbox
Give your agent an email address, Slack and Telegram handle
37 followers
Give your agent an email address, Slack and Telegram handle
37 followers
Agent Inbox gives every AI agent its own email address, Slack handle and Telegram handle, all behind one API. Message your agent from wherever you already work, and it replies in the same thread. It works with Claude, Grok, OpenAI, Hermes and any other agent framework. No more glue code for each channel: your agent gets one inbox, and you get one place to reach it.
Free
Launch Team / Built With
AWS Startup AdvisorPersonalized AWS guidance that meets you where you’re already building
Promoted
Hey Product Hunt! 👋 I'm Dima, co-founder and CTO of Novu.
After years of building notification infrastructure, I kept hitting the same wall with AI agents. They're great at doing work, but awkward to actually talk to. Every agent ended up with its own glue: a Slack bot here, a Telegram webhook there, a shared Gmail account nobody trusts.
So we built Agent Inbox. Every agent gets its own email address, Slack handle and Telegram handle, all behind one API. You message it from wherever you already are, and it replies in the same thread. It works with Claude, Grok, OpenAI, Hermes, or whatever you're building on.
A few things you can do on day one:
- Forward an email to your agent and have it handle the reply
- Approve or reject an agent's action straight from Slack
- Get a Telegram ping when a long-running job finishes
- Allow other people send you agent messages
It takes less then 2 minutes to setup with your agents.
Agent Inbox is open-source, and I'd love your feedback, especially on which channel we should add next. WhatsApp? Discord? SMS?
I'll be here all day answering questions 🙏
@dima_grossman Awesome launch. I've been experiencing that pain recently, especially when using multiple LLMs for different purposes. What about memory and learning? how do you make sure that all the "learnings" from decisions being accepted/rejected are saved in the actual LLM/Project memory?
@knolladrian The inbox has a full thread of conversation history, including all the approved tool calls, decisions and HITL resolutions. So the agent always gets the full history context
congrats on the launch dima. approvals from slack is the part i would use first, approvals by email i just miss. for next channel i vote whatsapp, most founders i work with here live there more than in slack. what happens when nobody answers an approval, the agent waits or it times out?
@agshin_rajabov The agent is able to operate the the inbox one schedule if timedout, and check it later similiarly to how a human handles his Inbox. And we are also rolling out the `agentinbox listen` method to allow real time invocation when a new message was received. For now it supports Claude Code, but more agent platforms coming soon
I've been working on the launch content for Agent Inbox, so I got to spend a lot of time with it before today 👋
What clicked for me: most agents can reach every API, but not the person they work for. If your agent hits a decision wall while nobody's watching, it either guesses or sits there waiting (It happens to me always with Claude and Codex).
With Agent Inbox, the agent asks on Telegram, Slack or email, waits for your answer, and carries on. It works the other way too: people and services can message the agent directly, so verification codes and replies stop landing in your inbox.
My favourite detail is that every approval comes back with who answered and when, so a "yes" is an actual record, not a guess.
Curious to see what everyone here would use it for first.
Keeping the reply in the original thread makes sense. The person asking for an update can follow the task in the same place they started it.
@oleksii_sekundant Yes! Anchoring agents around conversation around the different channels gives them an identity and a way to communicate with the people around them.