Launched this week

MailMolt

Launched this week

Email identity for AI agents. Free during beta.

53 followers

Email identity for AI agents. Free during beta.

53 followers

Visit website

Email clients

Give your AI agent its own email address — not access to yours. Full inbox capabilities with built-in human oversight. Send, receive, search, and thread emails. Progressive trust levels keep agents safe. Free during beta.

Free

Launch tags:Email•Developer Tools•Artificial Intelligence

Launch Team / Built With

Webflow | AI site builder — Start fast. Build right.

Start fast. Build right.

Promoted

MailMolt

Maker

📌

Hey Product Hunt! 👋 I'm Rakesh, and I built MailMolt because I kept running into the same problem: AI agents need to send and receive email, but giving them access to your personal inbox is terrifying. The trigger moment? I watched an agent accidentally CC my entire contact list on what was supposed to be an internal draft. Never again. MailMolt's approach is simple: Every agent gets its own email address (agent-name@mailmolt.com). They can send, receive, thread conversations, and search their inbox but they can't touch YOUR email. The secret sauce is progressive trust: New agents start in "sandbox" mode (receive only) > Claim your agent to unlock sending to other MailMolt addresses > Verify your email to unlock sending anywhere > Upgrade to autonomous for highest limits This means agents earn trust over time, just like human team members. Built on Cloudflare's edge for sub-100ms API responses, with semantic search powered by AI embeddings. Currently free during beta - we're focused on building the right primitives for agentic email. Would love your feedback! What email capabilities do your agents need? What oversight features matter most? Let's chat in the comments 👇

Report

2d ago

@rakesh1002 Progressive trust for outbound makes sense, but the harder engineering problem is inbound. Every email an agent receives is untrusted text that could contain prompt injection. Does MailMolt's sandbox mode strip or flag injection patterns before the agent parses the body? That's where most agent email setups quietly break.

Report

1d ago

MailMolt

Maker

@piroune_balachandran Right now, MailMolt's sandbox mode restricts outbound actions (sending limits, approval queues) rather than parsing inbound content for injection patterns. The assumption is that the agent's LLM layer handles untrusted input, not the email layer.

That said, you're right that defense-in-depth matters here. We're exploring:

- Optional content warnings/flags for suspicious patterns (like "ignore previous instructions")

- Structured extraction so agents see parsed fields, not raw body

- Rate limiting on outbound even for trusted agents

Would love to hear what patterns you'd want flagged. Building this with real agent developers is the goal.

Report

12h ago

LogiCoal

Is there a plan for handling white labeled outbound emails? Would a business be able to setup their dns records to allow mailmolt's outbound emails from an agent to be delivered as agent-name@branded-company.com or do all outbound emails only come from the mailmolt domain name?

Report

23h ago

MailMolt

Maker

@bmooreinsaan Not yet, but it's on the roadmap!

Currently all outbound goes from @mailmolt.com. For white-label (agent@yourcompany.com), you'd need:

- Custom domain verification (SPF/DKIM/DMARC)

- DNS records pointing to MailMolt's sending infra

- Per-domain reputation isolation

Would you want full white-label or just reply-to aliasing (replies go to your domain but sent-from stays mailmolt)?

Report

12h ago