Trust & data

A gateway sits between you and the model, so it’s worth knowing exactly what it does with your data and where the models come from. Here is the whole policy, in plain language.

We don’t store your prompts

The gateway meters tokens, not content. Each usage record holds counts, the model that answered, timing, and a routing signal vector — never the text of your prompts or completions. There is no prompt log to leak, subpoena, or train on, because it doesn’t exist.

Compliant model sourcing

Open-weight Chinese models are the price leaders, and we serve them — carefully. GLM is routed through US-based inference hosts rather than a single origin, which is the industry-standard, compliance-aware way to serve it. We keep a diverse catalog (GLM, DeepSeek, Qwen) so no single supplier is a point of failure, and we remove any model whose supply or compliance status is unsettled — which is why Kimi is not currently in the catalog.

If a sourcing question ever affects availability, we’d rather drop a lane than quietly serve something we shouldn’t. Catalog changes are visible on the models page.

The measured-savings pledge

The AI-cost-tooling market is full of invented “tokens saved” counters. We don’t do that. Every savings figure we show is computed from the metering ledger for your actual traffic, and our routing methodology is public. If a number can’t be measured, we don’t display it.

Payments

Balances are prepaid and processed by the payment provider you choose (bKash, Nagad, UPI, or card) — we don’t store card or wallet credentials. Consumed tokens are non-refundable; unused balance stays on your account. Prices are shown tax-inclusive where local rules require it.

See pricing Set up your tool