v1.2.0 — PDF support & language hints

Multi-page PDF → text, now with language hints and PyPI-ready packaging.

New

• PDF → text: pages rendered and OCR'd one-by-one, labelled [Page N], up to 25 pages / 12 MB
• Language hints: optional lang param (en, ch, japan, korea, …) in API, CLI and UI
• PyPI-ready: proper wheel/sdist packaging, web assets shipped as package data

Engine & limits

• Images: PNG / JPG / WEBP / BMP / TIFF — up to 12 MB
• PDFs: up to 12 MB, 25 pages (free-tier runs ~30–60 s/page)

See the README for the live demo and full API reference.

textrieve v1.1.0

Image to text — free, fully open. Auto-expiry + free-tier hardening.

What's new

• 2-min auto-expiry: extracted text is wiped from the page automatically (live countdown)
• Free-tier memory guards: image released + GC per inference, 4 MPx cap, concurrency limiter (2 parallel ONNX passes, bounded queue, 503 under load)
• Robust API: unreadable images -> 400; clear 413/415 errors
• UI polish: favicon (fixes dead-image 404), Open Graph tags, privacy note, countdown pill
• Deploy: Dockerfile + render.yaml for Render free tier; static front-end fits Vercel
• Sponsor: FUNDING.yml + sponsor button

Stack

FastAPI + RapidOCR (ONNX, CPU) + Pillow. No API keys, nothing stored server-side.

Tests

10/10 (real OCR round-trip, blank/corrupt/oversized uploads, UI+favicon). CI on 3.11/3.12.

textrieve v1.0.0

Free, fully-open image-to-text (OCR).

• Web UI: drag-and-drop, preview, copy/download, engine status
• REST API: POST /api/ocr (multipart) -> text, confidence, duration
• CLI: single file, many files, or a folder; plain or --json
• Engine: RapidOCR (ONNX) on CPU — offline, no API keys
• 7 tests incl. real OCR round-trip; CI gate (Python 3.11/3.12)
• Apache-2.0