What we see, why we see it, and what we do with it

Our product is clarity on your inference

Valni's core product is giving you data about your inference: what it costs, how much you use, and who is using it, across every provider. Whether you are a solo engineer, a team, or a SaaS company metering your own customers, that is what you come to us for.

We are a true proxy and inference gateway. All of your traffic travels through us on its way to the model providers. That is the point, and it is what makes the features powerful: routing, one set of provider keys and accounts you never touch, spend limits, and exact per-user cost attribution all depend on us sitting in the path.

That access is a responsibility, and the line is simple: we never sell, save, analyse, or train on the content of your messages, full stop. Our mission is transparency, both about your inference use and about how we deliver it. This page is part of that. It tells you exactly what passes through us, what we keep, and what we never touch.

Measuring is the job, not surveillance

The reason we see the traffic is accounting. Attributing an exact cost to every key, every team member, and every provider is the core of what Valni does, and you cannot get a per-person ledger without measuring per-person usage. So we measure every request: the tokens, the price, the model and provider, a timestamp, and who made the call. That is the ledger you are paying for. It is a record of what a request cost, not a copy of what it contained.

What we store

We keep those measurements. We do not store your prompts, and we do not store your model's responses. The record is what a request cost and who made it, never what it said. Your conversations are not ours to hold, and we do not hold them.

Permission controls you turn on

Our permissions policy tool runs server-side when you turn it on. In a response, the tool calls are a separate, structured field, apart from the message text. The tool reads only that field, never your messages, and matches your rules against the tool calls your agent wants to make, the commands and file paths and network calls, allowing, gating, or blocking each one. It keeps nothing. If you set no policy, nothing is read.

What you control

Routing, billing, and spend limits never read your content; they work from the measurements above, like which model you called and how many tokens it used. Everything is opt-in, so the fewer features you use, the less we touch. You still get routing, cost attribution, and spend controls, and nothing you send or receive is ever read.

What we never do with your messages

We never sell, save, analyse, or train on the content of your messages. The only data we collect and use is secondary. It is the measurements that produce your ledger and your controls, never the messages themselves.

Passing through is not collecting

Because we are a proxy, your data passes through Valni to reach the provider. Passing through is not the same as keeping. We collect only what the features you use require, mainly the measurements behind your ledger and spend controls. Use fewer features and we hold less. We do not collect anything for its own sake.