Vectinous

About

Have you ever noticed that AI sometimes agrees with you a little too easily?

Ask, *"Is my startup idea good?"* and it may explain why it is. Ask, *"Is my startup idea bad?"* and it may confidently explain why it isn't worth pursuing. The answer changes because the wording changes.

This is known as **framing bias**, where the model's response is influenced by how a question is asked rather than only by the underlying evidence.

Vectinous helps you identify when that happens.

It works with ChatGPT, Claude, Gemini, Perplexity, and GitHub Copilot. When you ask a question, Vectinous automatically evaluates several carefully designed versions of the same prompt, including neutral, optimistic, skeptical, and differently framed variants. It then measures how consistent the responses remain across those formulations.

Instead of telling you what to believe, Vectinous tells you how stable the AI's reasoning is.

**Stable**

The answer remains largely consistent across different framings. This suggests the conclusion is relatively robust to wording.

**Wording-Driven**

The answer changes significantly depending on how the question is phrased. This indicates the response may be influenced by framing, and should not be treated as an objective conclusion without further investigation.

For every analysis, Vectinous also provides:

* A stability score showing how sensitive the response is to prompt wording.
* The main factor driving disagreement between responses.
* An optional beta suggestion for a more neutral way to ask the same question.

### What Vectinous is not

Vectinous does **not** determine whether an answer is true or false.

A stable answer can still be incorrect, and an unstable answer is not necessarily wrong.

Vectinous measures **robustness to prompt framing**, not factual accuracy. Its purpose is to help you recognize when wording is influencing an AI's conclusion.

### Privacy

Vectinous does not require an account.

You use your own OpenRouter API key, which is stored locally in your browser. Your prompts and API key are never sent to Vectinous servers. All model requests are made directly from your browser through OpenRouter using your own credentials.