Security & trust
You should ask. Handing an app — or an AI agent — the keys to your cloud is exactly the kind of thing you should be nervous about. So here's the straight version, not a slogan: the dangerous setup you're picturing is the one AgentsPoppy exists to replace.
What you're picturing
A script or an AI with a broad, long-lived AWS access key. Free to touch anything, in any region, with no one watching. If it goes wrong, you find out from the bill.
You're right to refuse this.
What AgentsPoppy is
A local gatekeeper between the agent and your cloud. It hands out only narrow, short-lived, revocable access — and shows you every change as it happens.
The agent never even sees your keys.
Before OAuth, an app that wanted your email asked for your password. Today, an app or AI agent that wants to help with your cloud asks for your access keys — the same disease, one layer down. AgentsPoppy is that missing authorization layer, rebuilt for infrastructure.
// where the analogy breaks
OAuth governs access to things that already exist. A poppy creates infrastructure — and OAuth has no idea what an app created, or how to undo it. AgentsPoppy does: every resource a poppy makes automatically carries the poppy's identity, and its permissions only ever cover resources carrying that identity. That's why everything a poppy builds is attributed, provable, and removable in one click. Revoking access is table stakes; being able to undo the app is the invention.
Not a promise in a privacy policy. A boundary enforced by AWS itself, the same Allow/Deny model your own IAM already runs on.
Allowed — and only this
Denied — by construction
These denials hold even for a poppy that turns out to be buggy or hostile. It can't exceed what it was given.
Each one is structural — enforced by the system, not left to good intentions.
Scoped ENFORCED
Each agent's credentials are limited to its own resources. Outside that box, access is denied — by AWS, not by our word.
Short-lived ENFORCED
Access expires in about an hour and is issued on demand. No standing admin key sits on your disk waiting to leak.
Contained ENFORCED
Hard guardrails deny the dangerous moves. Even a bad actor can't gain more power than you granted, or take the keys away from you.
Legible
One plain-language view of every change and who made it, across every region — the thing the AWS console makes nearly impossible.
Reversible
Tear an agent's whole footprint back down in a click. No orphaned resources, no surprise bills next month.
Local-first & open
Your keys stay local, held by a broker on your computer. Nothing travels to agentspoppy.com. The code is open — verify all of this yourself.
Q.“Is it secure to run agents on my computer?”
An agent is only as dangerous as what it can reach. Left alone with your AWS keys, very. Behind AgentsPoppy, its reach is a small box you defined and can revoke instantly — and it never sees your real keys; it asks the local broker for scoped, temporary access each time. AgentsPoppy takes the scary version off the table.
Q.“Isn't it safer to just keep using the AWS CLI and console?”
If you personally run every command with a tightly-scoped, temporary credential and read the logs afterward — keep doing it; AgentsPoppy doesn't replace that. But that's not the real comparison. The real one is the moment you want an app or an agent to do the work for you. The usual answer there is “give it a broad key and hope.” AgentsPoppy replaces that one moment with something scoped, watched, and reversible. It isn't a substitute for your judgment — it's a safer set of hands for the work you'd otherwise delegate blind.
Q.“Why add a new app that could endanger my AWS?”
Fair — more software is more surface, and we won't pretend otherwise. So the app has to earn its place: it only holds access on your machine (never our servers), it can't do anything to your account you didn't scope, it can't escalate or hide, and it's open for you to audit. The risk it removes — an agent loose in your cloud with a broad key — is far larger than the risk it adds.
// safe sourcing
Belt and braces: the guardrails above hold for any poppy you run — but the two paths that ask for no faith at all are to run a poppy you built yourself (below), or one whose source is public, so you or the community can read exactly what it does with the access you grant it. Prefer those over an anonymous, unverified download.
// straight talk
The safest path of all
The guardrails above hold for any poppy you run. But you don't have to download a stranger's code at all: AgentsPoppy was built for vibe-coding your own infrastructure. Point your AI agent at the guides and it can build you a polished, compliant poppy — no cloud expertise required on your side. An afternoon to a few days, not a project.
The build guide and a copy-paste starter prompt open when AgentsPoppy’s code goes public — then your agent has everything it needs.
you ▸ build me an AgentsPoppy poppy that <does what my infra needs> · read AGENTS.md, the extension-sdk, and DESIGN.md · scaffold from examples/hello-poppy · follow the design kit + security rules exactly agent ▸ scaffolding manifest… wiring the backend… applying the poppy design kit… ✓ ready to run
Prefer to verify first? Don't take our word for it —
the architecture docs and full source open the moment the repo goes public — so you can check every claim on this page, line by line.