I BUILT LUCENTASO YOU DON'T HAVE TO

Every product with a comment box, a chat feature, or a signup form eventually has to answer the same question: what happens when someone submits something toxic, abusive, or full of personal data that shouldn't be stored? Most teams either bolt on a slow, expensive enterprise moderation contract, or hand-roll a keyword blocklist that a five-second workaround defeats.

I wanted something in between — a content moderation API a solo developer could integrate in an afternoon, fast enough to run on the request path instead of an async queue, and honest about what it does and doesn't catch.

Lucenta runs entirely on Cloudflare's edge network. The profanity and toxicity checker embeds incoming text with a Workers AI model and searches a Vectorize index for the closest semantic match, so it catches paraphrased and lightly obfuscated abuse, not just an exact string match. The PII redactor is pattern- and checksum-based, including Luhn validation for card numbers, which is why it responds in single-digit milliseconds with no model call at all.

There's no central server to route through, no cold start, and no queue to wait behind — the same infrastructure decisions I'd make if this were my own product's moderation layer, because in a sense, it is.

Speed, because moderation that lags behind your users is moderation that ships toxic content anyway. Privacy, because text submitted to the API is checked in memory and never written to a database or log. And a developer experience I'd actually want to use myself: one API call, clear responses, no SDK to install.

Multi-language detection, a granular toxicity score instead of a binary flag, and support for custom word lists are next on the roadmap — all covered in more detail on the Learn page.