PromptGate – Prompt regression bench

Run regression tests and safety guardrails on prompts right in the browser

Problem

Before and after launch, a single prompt edit can skew outputs or leak sensitive content. Developers test ad hoc across scattered scripts with no unit-test-like regression, and while some browser-native prompt QA exists, most teams still lack a lightweight way to wire it into CI.

Solution

Maintain a suite of test cases in the browser (inputs, expected behavior, banned words, jailbreak instructions), run a one-click regression on each prompt change, and get pass rates, risk items, and diffs. Export results as CI-callable JSON and surface suggestions for blocking overreach instructions.

Users

Small teams shipping AI features, indie developers, PMs who review prompt changes

MVP

Case set management with versioning; one-click regression with pass rate; sensitive-word and jailbreak detection; diff and JSON export; simple CI API

Revenue

Free tier with 200 runs/mo; Pro subscription (~$12/mo) for CI API, team collaboration, and historical trends

Why now

Prompts have effectively become code but regression and safety tooling is blank; front-end plus a thin proxy is enough to validate without heavy compute