AIQualityHQ

AIQualityHQ is a free, browser-native toolkit for AI developers, QA teams, and prompt engineers. It allows users to score prompts across 6 core dimensions, refactor raw instructions for specific LLMs, optimize token costs, and scan for prompt injection and jailbreak threats using a deterministic engine that runs entirely client-side for complete privacy and instant execution.

The seven browser tools are free with no card and no account, and the free tier is described as free forever. Everything interactive runs client side, so browser use is not metered at all.

What the paid tiers sell is the same engine over a REST API. Free includes sixty API requests an hour. Pro is $19 a month for ten thousand requests an hour, the SDKs, unlimited report history, the AI visibility scorecard and priority email support. Team is $49 a month for fifty thousand requests an hour, shared workspaces, SSO with role-based access, custom rate limits and dedicated support. Paying yearly takes twenty percent off.

API limits reset every hour and are shared across the checker, scanner, detector and audit endpoints. Browser tools never count against the quota.

So the decision is narrow: stay free unless you want to call the engine from CI or a backend, at which point Pro is the first tier that has an API at all.

AIQualityHQ suits people shipping LLM features who want a check before production: prompt engineers validating instructions, QA engineers testing whether behaviour holds, and developers looking at injection risk.

The local-first design is the reason to pick it over a grader that calls a model. Prompt text stays on the machine, results are reproducible, and there is no key to manage or quota to watch. That also sets the limit. A rules engine reads structure and known patterns; it does not judge whether your prompt gets a good answer from a given model, and it cannot tell you that your instructions are wrong about your own product.

Use it as a fast, repeatable first pass. Judgement about output quality still needs an eval that actually runs the model.