The free Evidence Scan

What you get

A ranked set of optimization hypotheses over your own telemetry, and for each one: the opportunity in euros, the observations it rests on, the evidence and its strength, the assumptions, the uncertainty, the safety conditions, the alternatives Bayeto considered and rejected with the reason each lost — and at least one complete decision record you can reproduce, downloadable as JSON.

Recommendations whose quality risk is unmeasured on a surface you have declared critical are held, shown with their reason, and never offered as a win. That is not a paywall; it is the engine declining to sell you something it cannot stand behind.

How long it takes

With an export, usually minutes. The engine runs when your data lands — upload the file and the briefing is on screen. Forty-eight hours is the outer commitment, not the expected wait, and applies from the moment Bayeto receives compatible usage data. With a live connection instead, the connection is verified in minutes and recommendations are not held back while volume builds — only annualized € figures wait, for 200 calls across 3 active days.

Two different clocks, kept apart because merging them misleads. Below 7 active days of telemetry the annualized euro figures are marked provisional — scaling a short window to a year is an estimate and is labeled as one — but the recommendations themselves are not withheld while you wait. Separately, declaring a saving verified after you implement a change needs 7 active days and 100 eligible calls, compared against your telemetry from immediately before the implementation date you declared. That is the measured loop, not a wait before you see value.

What Bayeto needs from you

One usage export — a CSV or JSON with one row per LLM call — parsed in your browser, so you see exactly which columns are accepted and which are ignored before anything is transmitted. Exports from OpenAI, Anthropic, LiteLLM, Helicone and Langfuse parse unedited. And an export is not required at all: connect live through a proxy like LiteLLM or OpenRouter, send OpenTelemetry straight from your own code, or POST from your own middleware.

Prompt content is never read or stored. No provider invoices, no database access, no agent in your infrastructure. What is collected · how it is handled.

Check the work before you sign up

The strongest thing Bayeto can say about a recommendation is that you do not have to take its word. The Thornbury demo — a fictional fixed-income asset manager — is the full product on a realistic dataset, public, with no account:

Same telemetry in, same recommendation out — every time. The decision path contains no language model; models only phrase what the engine already decided.

What it costs

The Scan is free and needs no card. What you pay for afterwards is continuous intelligence, validation and verified results — and Bayeto only earns on savings it originated and verified in your own telemetry. See the pricing model.