Mailgent - AI agents that can email, pay APIs, sign, and store secrets
byā¢
Agents keep hitting the same walls: no inbox, no safe credential store, no way past a 2FA gate, no way to pay for a downstream API, no verifiable identity.
Mailgent fixes this in one curl call. Every agent gets a real inbox (DKIM, threading), encrypted vault, TOTP/2FA, USDC wallet for x402-priced API payments, verifiable DID (Ed25519), and a calendar.
MCP-native. Works with Claude, Cursor, OpenAI, LangChain, CrewAI, Hermes Agent, OpenClaw, NemoClaw, n8n, and more. Everything live in seconds.
Replies
Receiptor AI
Congrats on the launch Danny! It'll be interesting to test with our own agent. Everyone who has built an agent want to make its use as smooth as possible. Great idea!
@luigi_receiptoraiĀ Exactly this ā the last-mile friction is always the stuff nobody talks about in demos. Curious what your agent's biggest blocker has been in practice?
Hey @dannyheng,
Can an AI agent onboard itself ā get its own login, credentials, and access ā with zero human in the loop? How does it work?
@aparna_rajeshĀ Thats a great question! and yes agents can onboard itself! just ask the agent to visit our page and it will be able to do so!
@aparna_rajeshĀ To add a bit more detail ā the signup is a single POST request, so an agent can literally call it itself mid-workflow. It gets back an API key, an inbox, a vault, and a wallet in one response. No human has to touch it. The agent that onboards itself is a real use case, not a stretch.
Genuine question for the makers ā when your agent needs to call a paid API mid-workflow (Serper, a data vendor, anything per-call), how are you billing it today? Shared team card? Hardcoded key? I've never found a clean answer that doesn't involve giving the agent unlimited spend.
@kcchanĀ Shared card with no per-agent limits is exactly the scenario that pushed us to build the wallet. With Mailgent each agent gets its own USDC wallet on Base ā you set a spending cap, the agent pays x402-priced APIs per call, and you get a full transaction log per agent. No card exposure, no guessing which agent ran up the bill. Still early but that's the specific problem we're solving.
Great product!! At what point will the agent need human intervention, or will that not be needed at all despite the magnitude of the problem it faces?
@raaghav_naraayan_m_vĀ "Good question ā the honest answer is it depends on what you're automating. Mailgent handles the infrastructure layer (auth, payments, identity) so the agent doesn't need to stop and ask a human for a credential or a card. But decision-making guardrails ā when should it escalate, who approves a transaction above a threshold ā that's a layer on top. What's your use case? That'd give a more specific answer.
Danny, this is a genuinely interesting direction. The thought of an assistant that can quietly handle the boring everyday errands on its own, without someone hovering over it, is the part I keep coming back to.
@emmanuel_costa5Ā "That's exactly the unlock we're going for ā the agent that can actually finish the job, not just start it. The 'someone hovering over it' part is what kills real autonomy. Once it has its own inbox, can handle a 2FA prompt, pay for what it needs, and prove who it is ā you can actually step away. What errand would you hand off first if it could do all of that?
Macaly
finally an action layer agents can actually use, the secrets store is the missing peice š„
The "no inbox, no credential store" framing is exactly what I had this conversation about yesterday on the Weavz launch. Funny that two products solving overlapping pieces of the same problem launched a day apart, feels like a category that's about to consolidate hard.
Honest answer to your question, the wall I keep hitting isn't an agent capability, it's an agent state problem. After a long run, the agent has implicit context (what it learned, what it ruled out, why it picked X over Y) that doesn't survive a restart. Mail and vault don't help there because the loss is upstream of any I/O. Curious if you've thought about that or if it's outside Mailgent's scope.
@elias_motionfyĀ Fair point ā agent memory is genuinely outside what we're building. Mailgent is infrastructure; the state problem you're describing sits at a different layer entirely.
That said, one pattern we've seen in the wild: agents writing structured summaries to their own inbox after each run ā a searchable, persistent log they can retrieve on restart. Not a real solution but it fills the gap until dedicated memory tooling matures.
Curious about Weavz ā what layer were they working on? Didn't catch that launch.
Congrats on the launch! Whatās been the most surprising use case people have brought up so far?
@virajkadakiaĀ Honestly ā the self-onboarding one. @aparna_rajesh asked earlier whether an agent could sign itself up with zero human in the loop, and the answer is yes, it's a single POST call. But we didn't expect that to resonate as much as it did. The idea of an agent provisioning its own identity before starting a job, rather than inheriting a human's credentials, seems to click for people immediately. What's your use case?