Launching today

Puffin
Identity and inboxes for AI agents
55 followers
Identity and inboxes for AI agents
55 followers
Give each AI agent its own email address, a seat in Discord, Slack and Telegram, and a calendar. It replies with built-in AI or your own agent: you approve every message, or let it run on full autopilot.





Payment Required
Launch Team / Built With

TinesThe single, secure environment for agents, apps, and automations.
Promoted



Hey there, fellow Hunters!, I'm Divyanshu. @akarsh_kushwaha and I built Puffin, and this is our first launch.
We kept running into the same wall with AI agents. They can write code, research and draft replies, but they can't actually reach anyone. They have no email address, no place in the team's Slack or Discord, no calendar. Every time an agent needed to talk to a real person, we ended up copying text around by hand or giving it access to our own inbox and hoping for the best.
Puffin gives your agent its own identity. It gets a real email address and can work with Gmail and Google Calendar. It can sit in your Slack, Discord or Telegram and answer people there. It replies using Puffin's built-in AI or your own agent, and nothing real goes out until you approve it, either in the dashboard or with one click from an email. We also added the guardrails we wanted for ourselves: a hold on messages that look like prompt injection, virus scanning for attachments, sender rules and daily limits.
The part we use most ourselves is coding from the phone. You message your agent on Telegram or Slack, something like "fix the failing login test", and Claude Code or Codex works on your own machine. Before every edit or command, it asks you in the chat to allow or reject. It means you can step away from your desk without stopping the work.
The plugins for Claude Code, Codex, Hermes and OpenClaw are open source on our GitHub (github.com/justpuffin). If you'd rather bring your own agent, there's a hosted MCP server with 44 tools, plus Python and JavaScript SDKs.
We're part of NVIDIA Inception, AWS Activate and Claude for Startups. A few YC founders tried early versions, and their feedback shaped a lot of what's here. Phone numbers, SMS and iMessage are what we're working on next. Plans start at $25 a month.
I'd really like to know: what's the first job you'd give an agent that can message you back? We'll be here all day answering every comment.
Get 25% off for first 100 fellow builders - Use Discount Code - RBDIRI1HKCDN
Really cool project! What are the limitation so far in term of access or generally speaking?
@thomas_wazza On access and data: we've submitted our Google verification request for Gmail and Calendar, and our MCP server for Anthropic's review. Until Google approves, Google shows an "unverified app" screen when you connect Gmail or Calendar, which you can click through.
We only store what's needed to run the features you turn on, such as your agent's messages and contacts. We never sell it, use it for ads or use it to train AI, and you can export or delete your workspace at any time. Our privacy policy has the details: justpuffin.tech/privacy
Can an agent check when I’m free without seeing what is on my calendar?
@viktoriia_vasylchenko That's how the calendar integration is designed: the availability check returns only busy times, not event titles or details. Google Calendar is still rolling out, so it isn't open to everyone yet.
Hey Product Hunt, I'm Akarsh, the other co-founder of Puffin. @divyanshurajput told the story, so I'll cover how it works under the hood.
The thing we cared about most is that an agent should never surprise you. Every email, meeting invite and code change waits for your approval. Incoming mail is checked for prompt injection and held if it looks suspicious, and every workspace has daily send limits and an audit log. Replies inside a chat you connect, like Slack, Discord or Telegram, are automatic by default, and you can switch that per chat.
For developers, there's a hosted MCP server with 44 tools, Python and JavaScript SDKs, a CLI, and open-source plugins at github.com/justpuffin. The Claude Code one is the most polished, and the Codex, Hermes and OpenClaw ones are early. You can run Claude Code on your own machine and approve each edit or command from Telegram. It runs on your computer, not ours.
The hardest part for me was letting several agents share one chat without talking over each other. So every chat has a default agent, and people reach the others by starting a message with their name.
It's early, so please tell us what's missing. Phone numbers, SMS and iMessage are next. I'll be in the comments all day, and I'd love to hear what you'd build with it.
giving each agent a separate inbox makes it easier to see which one sent a reply. i'd want an approval queue that shows the sender, recipient and channel together.
@saad_el_gueddari Totally agree, and that's how we built it. Every agent has its own identity and address, so you always see which one is talking. The Approvals queue shows each request with the sending agent, the recipient and the channel side by side, plus the full message, and you can approve or reject in bulk. You can also approve right from Telegram or Slack, or with one click from an email alert.
If something in that view is missing for how you'd work, tell me. I'd love to hear it.
The approval-before-send flow is what makes this click for me — I wouldn't hand an agent a real inbox without it. The per-agent identity is a smart way to frame it too: the agent sending as itself instead of as me changes the whole trust question. First job I'd give it: watch my inbox for App Store review mail and ping me on Telegram the moment something lands, instead of me checking five times a day.
@jcbuilds yes that's the the problem which is resolved by Puffin; please try, build your first agent and inbox for it at www.justpuffin.tech
looks cool!
@louislecat thanks
@louislecat Thankyou!