Every API platform eventually gets a "catch-all" namespace for the utilities that don't belong anywhere else. We resisted that framing for /tools. Instead of a junk drawer, we designed it as a coherent developer toolkit — 20 endpoints across five categories that cover the repetitive infrastructure work that shows up in almost every SaaS product.
Why these 20 endpoints?
We looked at what developers building on the other nine namespaces actually needed next. The patterns were consistent: parse a document (OCR, PDF split), process media (compress an image, transcribe a recording), validate user input (email, phone), do NLP on text (translate, summarise, sentiment). These are the endpoints that every product team ends up stitching together from four different providers, each with its own key, rate limit, and billing cycle.
Consolidating them under one key isn't just convenient — it changes the economics. A startup building an invoice processor shouldn't need to budget separately for an OCR service, a PDF library, and an email validation provider before they've validated the idea.
AI-powered vs utility endpoints
The 20 endpoints split into two tiers internally. Utility endpoints — QR code generation, image compress/convert/resize, URL shortener, PDF merge/split/compress, email and phone validation — are pure computation with no upstream AI dependency. They draw only from your plan's regular monthly quota.
AI-powered endpoints draw from a separate, smaller AI quota that each plan allocates monthly. OCR, audio transcription, TTS, and resume parsing run on GPT-4o Vision and Whisper — the most capable (and most expensive) models — and draw from the AI quota at a 1:1 rate. Text summarisation, translation, sentiment analysis, and language detection run on GPT-4o-mini and draw from a separate lighter AI quota: 2,000 calls/month on Starter, 10,000 on Growth, 50,000 on Scale. The two AI pools are independent of each other and independent of the regular quota.
OCR — why GPT-4o vision instead of Tesseract
The first instinct for OCR is always Tesseract. It's free, runs locally, and has been around for decades. We tried it. The accuracy on anything other than clean, horizontal, high-contrast text is poor — scanned receipts, angled photos of documents, handwriting — all fall apart. GPT-4o vision handles these cases naturally. The trade-off is cost and latency, which is why OCR draws from the AI quota rather than the regular quota.
The endpoint accepts both multipart file uploads (field name image) and JSON with an image_url field, so it works equally well from a server pipeline or a browser-based tool.
PDF processing — pure Go via UniPDF
PDF merge, split, and compress are all handled server-side with no external API dependency. Merge accepts 2–20 files and an optional order parameter (comma-separated indices) so callers control page sequence. Split supports three modes: individual pages (?pages=1,3,5), a range (?start=1&end=3), or a fixed stride (?every=2). Compress does lossless cross-reference table optimisation — not image downsampling — so the output is byte-for-byte identical in content, just with leaner internal structure.
All three return binary PDF data with custom headers (X-Total-Pages, X-Saved-Percent, etc.) so callers can track processing metadata without parsing the body.
Image utilities — Lanczos3 resize, not AI upscaling
The image namespace covers compress, convert, resize, and metadata extraction. Resize uses Lanczos3 pixel resampling with optional unsharp mask — it's deterministic, fast, and correct up to 8000px output. We explicitly did not call it "AI upscaling" because it isn't. The X-Method: algorithmic-lanczos3 response header documents this.
Metadata extraction returns dimensions, DPI, colorspace, alpha channel presence, dominant colour (RGB + hex), per-channel histogram statistics, and whether EXIF data is present. Useful for content pipelines that need to categorise or validate images before storing them.
Email and phone validation
Email validation does real MX record lookup, not just regex. The response includes a result field with one of four values: valid, disposable, risky, or invalid, plus a risk score and individual check breakdown. Disposable domain detection covers the major throwaway providers.
Phone validation parses and formats numbers in four formats (E.164, national, international, RFC3966), identifies country and line type (mobile, landline, VoIP), and returns risk flags. The country_hint parameter improves accuracy on ambiguous local formats.
URL shortener — owned infrastructure
Short URLs live at aplicious.com/s/[code] and redirect with a 302. Expiry is configurable (expires_in_days), and expired links return 410 Gone rather than silently redirecting anywhere. Owners get analytics via /tools/url/expand?code=.... Custom codes are supported — if the code is taken, the API returns a 409.
Two quota pools in one namespace
Having utility and AI endpoints in the same namespace required a decision: do we apply a single quota by namespace or track consumption by endpoint? We track by endpoint. Utility endpoints draw from the regular monthly quota; AI endpoints draw from the appropriate AI quota pool. The docs panel labels each endpoint's quota tier so callers know which pool a call will draw from before they make it.
What's next for /tools
The screenshot endpoint (full-page and viewport capture of any URL) is the most-requested addition. It's on the roadmap for Q1 2027. HTML-to-PDF is related but architecturally separate — it requires a headless browser rather than image capture. Both are planned but not yet scoped.
All 20 current endpoints are live. The free tier includes 250 regular calls per month across all namespaces; AI endpoints (OCR, transcription, TTS, resume parser, and the text AI tools) require a paid plan.
Revenue projections, starter code, and 6 business pitches — all on the /build page.