Most profanity filters are just word blacklists. They flag "Scunthorpe" and let "f/u" pass. The Profanity API runs a 5-layer pipeline ending with an LLM for intent classification — 300 free requests/month, then $0.002/request ($0.005 with the LLM layer).
Quick Reference
| Spec | Detail |
|---|---|
| Detection Layers | 5 (blocklist → dictionary → semantic word → semantic phrase → LLM) |
| Context Types | 12 recognized categories |
| Input Length | 3–1,250 characters |
| Authentication | Bearer token in Authorization header |
| Free Tier | 300 requests/month, 60 req/min rate limit |
| Paid Pricing | $0.002/base request, +$0.003 for LLM layer |
| Data Retention | 30 days (free tier) |
| Response Fields | Blocked flag, confidence score, classified intent |
| Customization | Custom blocklists + allowlists |
The 5-Layer Pipeline
- L0 — Custom blocklist: User-defined words and phrases. You control what’s always flagged.
- L1 — Static dictionary: Exact matches plus leetspeak variants ("a$$", "f**k").
- L2 — Semantic word matching: Catches misspellings and morphological variations.
- L3 — Semantic phrase-level: Understands context at the sentence level — distinguishes "this is fing good" (positive) from "f you" (attack).
- L4 — LLM intent classification: Handles edge cases, sarcasm, and ambiguous language.
The key distinction from simpler APIs (WebPurify, PurgoLink): layers L2–L4 interpret meaning, not just strings. This reduces the false-positive rate that plagues keyword-only filters.
Pricing Model
The pay-as-you-go structure means costs scale with usage. At $0.002/request, processing 100K comments costs $200. With the LLM layer active ($0.005 total), the same volume costs $500. The free tier’s 300 requests cover testing and low-traffic side projects. No subscriptions, no commitments — but also no refunds.
Known Issues
- Tiny footprint: Only 689 views and 9 saves since February 2026. The tool doesn’t appear in major "best content moderation API" comparisons (which list OpenAI, Google, Azure, AWS, Hive).
- Text-only: No image, video, or audio moderation. For multi-modal content, you’d need a separate tool like Hive Moderation or AWS Rekognition.
- 1,250-character limit: Each API call processes one text segment. Long forum posts or articles need to be chunked client-side.
- LLM layer cost doubles the price: At $0.005/request with LLM, the API is no longer cheap compared to OpenAI’s Moderation API (free for most use cases).
- No published accuracy benchmarks: The company claims context-aware detection but doesn’t publish false-positive/false-negative rates for independent evaluation.
- No Refunds policy: If the API doesn’t meet your needs after the free tier, you’re out whatever you’ve spent.
Visit The Profanity API — https://the-profanity-api.com/