Responsible AI · Puddles

This is how we build and run AI, whether it is a chatbot, an agent, an automation or a full application. The short version: you own the system, a person can always overrule it, your data stays where it should, and we tell you when AI is not the answer.

You own the system and your data

The code, the configuration, the accounts and the data are yours. We do not reuse your data for other clients, and we do not use it to train third-party models. When a project ends, the handover includes onboarding so your team can run and extend the work without us.

A person stays in control

AI drafts, people decide. We build features so a person can review the output, change it and switch it off. We do not hand consequential decisions to a model running unattended. Where a step is automated, we make the automation visible and reversible.

We show the limits, not just the wins

For every AI feature we build, you see what data it uses, what it does and where it can be wrong. Our demos run on fixed sample data for exactly this reason: we would rather show the mechanism honestly than fake a live result. We document what a system cannot do as clearly as what it can.

Your data stays where it should

We process data in the EU by default and grant access on a need-to-know basis, down to row level in the databases we build. Every third-party model or tool a solution depends on is named to you before it goes live. Client data is sent to a model provider only to produce your result, under that provider's data terms, and never to train their models. You choose the provider.

We scope to real value, and we say no

If AI is not the right tool for your problem, we say so on the first call. We would rather lose a sale than ship a feature that looks smart and does nothing. It is the same standard we hold our own products to.

We follow the EU AI Act

We track the EU AI Act and design for what it asks: transparency, human oversight and documentation, so your use of what we build stays defensible. Here is how the Act's transparency obligations (Article 50) show up in our own surfaces:

The demos on this site are replays of real AI runs, not live models. Every demo is labeled as a simulation, so no one mistakes it for a conversation with an AI.
Our AI-assisted communications go through human review before anything is sent, and Puddles Oy takes editorial responsibility for the content.
When we deliver a feature that interacts with users through AI, it ships with an AI-interaction notice and proper marking of outputs from day one.

None of our own uses fall into the Act's high-risk categories. This page describes our own practice and is not legal advice about your obligations.

Report a concern

If something in a system we built looks wrong, unsafe or unclear, write to puddles@puddles.fi. We respond without undue delay.