Stop building your own approval layer. Gate every agent action with Pliuz.

The problem now

AI agents aren't experiments anymore.

They're paying, refunding, emailing, and changing customer data in production. Recent incidents prove it: dropped production tables, a $31,000 payment with no approval, sensitive data exposed.

The question isn't “is governing agents worth it?”— it's “who answers when ours fails?”

Why not build it yourself

You could build the approval layer yourself.

And then you maintain it forever. Every one of these is its own project:

• State while a human decides
• Retries and timeouts
• Channel routing (Slack, email)
• An audit trail that actually holds up
• Row-level security per tenant
• Tamper-evidence you can prove

Pliuz is that layer as infrastructure — an open-source SDK and an EU-hosted control plane — so your team ships agents instead of approval plumbing.

How it works

From pip install to a defensible audit trail.

from pliuz import gated

@gated(risk="high")              # pauses for human approval in Slack
def refund(order_id: str, amount: int):
    stripe.refunds.create(order=order_id, amount=amount)

select ok, verified, total
from pliuz_verify_chain();   -- ok | 881 | 881  (tamper-evident)

$ pip install pliuz · Apache 2.0

Why you can trust it

Tamper-evident by design.

Every decision is a link in an append-only SHA-256 hash chain you can verify yourself with a single SQL call. EU-hosted in Frankfurt. DPA signable from day one. SDK open-source under Apache 2.0.

• EU · Frankfurt
• DPA day one
• Tamper-evident hash chain
• No LLM in the critical path
• Apache 2.0

Why design partners

Javier Mancera Villa

Co-founder & CEO

javier@pliuz.com

We're selecting 3–5 teams running 2+ agents in production. You get direct onboarding with me in the first two weeks, 40% off year one, preferred support, and real influence on the roadmap. Easy exit: 30 days' notice, your data exportable as signed JSONL.

Before you apply

The five questions that decide it.