WhAIQ · For agent platforms and orchestrators

Give your agents a calculation to call

Agents are good at running the steps of a process. When a step needs a number, like how much stock to reorder or which price to set, a language model generates it. WhAIQ calculates it on the model of the company, with the probability of hitting the target.

Schema: un agente porta avanti i passaggi di un processo, arriva al passaggio della decisione e chiama WhAIQ via MCP o API, che restituisce un numero calcolato come la quantità da riordinareDiagram: an agent runs the steps of a process, reaches a decision step and calls WhAIQ through MCP or API, which returns a calculated number such as a reorder quantity
Where it fits

One call, at the step that decides

1

Your agent reaches a decision

A reorder quantity, a price, a production plan, a marketing budget.

2

It calls WhAIQ

Your agent calls it through MCP or API, like any other tool, with no change to the framework.

3

WhAIQ calculates

Each alternative on the company's model, with the probability of reaching the target and the combinations ranked.

4

Your agent carries on

It continues the process with a number it can explain, or hands the choice to a person.

The evidence

The decision layer counts more than the model

In YC-Bench, a public simulation where an agent runs a start-up for a year and is scored on final cash, eight language models ran alone and with WhAI.

48 / 48

pairs better with WhAI: five models, paired test on four new worlds with no published runs

72% vs 5%

of the variance in final cash explained by the decision layer, and by the choice of model

Source: Grazioli et al., preprint on Preprints.org, 13 November 2026. Read the research

Inside your platform

It works as a tool your agents call

MCP

From your AI agents

Connect WhAIQ to the agents your platform already runs. Numbers calculated, not generated.

API

From your orchestration

Bring the calculation into the flows where decisions already happen.

Data

In Europe, per customer

Model and data run on Vedrai servers in Europe. Each client's model stays theirs, and their data never improve models used by other clients.

Security and data
Questions platforms ask

How it sits next to your agents

Do we have to change our agent framework?

WhAIQ is exposed through MCP and API, so your agents call it like any other tool.

What about small, frequent judgments, like routing a ticket?

Fast classifiers that return a label or a score fit those well. WhAIQ is for decisions whose answer is a number that moves a target: quantities, prices, budgets, plans. See how they compare

Who writes the result back?

Your agent. WhAIQ returns the calculation and your platform decides what to do with it.

How do we test it?

Request partner test access: you connect WhAIQ through MCP to your agents and try it on a case from one of your clients.
Read next

Integrate WhAIQ

For integrators who build agents for clients

Read

Decision layer compared

Where it fits next to other AI

Read

The Decision Layer

The research behind it

Read

Connect WhAIQ to one of your agents

Start with partner test access and a real case.