OCR drops Hindi/Devanagari and degrades romanized Hinglish — no Indic script support in Apple Vision

[ManmohanBuildsProducts]

Summary. On macOS, the screen→markdown OCR completely misses Devanagari (Hindi) text and partially garbles romanized Hinglish. This makes Familiar's screen-capture path largely unusable for Indian users, whose Slack/notes/work mix English, Roman Hinglish, and Devanagari.

Root cause. The OCR uses Apple Vision's VNRecognizeTextRequest. Its supported recognition languages contain no Indic scripts at all. Verified on macOS 26 (revision 3, accurate):

en-US, fr-FR, it-IT, de-DE, es-ES, pt-BR, zh-Hans, zh-Hant, yue-Hans, yue-Hant,
ko-KR, ja-JP, ru-RU, uk-UA, th-TH, vi-VN, ar-SA, ars-SA, tr-TR, id-ID, cs-CZ,
da-DK, nl-NL, no-NO, ms-MY, pl-PL, ro-RO, sv-SE

No hi-IN, no Devanagari/Tamil/Telugu/Bengali. So Devanagari on screen yields nothing or garbage. Romanized Hinglish is read as Latin but usesLanguageCorrection (English) distorts non-English tokens.

Compounding factor. Captures are downscaled to ~half display resolution (e.g. 960×540 from a 1920×1080 display), so even supported-language small text degrades. Larger on-screen fonts noticeably improve results, which points at capture resolution as a second lever.

Note: the clipboard-mirror path preserves full UTF-8 (Devanagari + emoji tested, byte-exact), so the gap is specifically the OCR pipeline.

Suggestions.

1. Allow configuring/raising capture resolution (or expose an OCR-fidelity setting).
2. For scripts Vision can't handle, fall back to a Devanagari-capable OCR engine (e.g. Tesseract with hin, or a configurable engine) — or at least let users add recognition-language hints.
3. Short term: document the Indic-script limitation so Indian users know to rely on the clipboard path.

Environment: macOS 26 (Tahoe), Familiar v0.0.70, Apple Silicon.

[Hyaxia]

Hey @ManmohanBuildsProducts thanks for the issue.
Its surprising to me that Hindi is not supported by apple's native OCR... that's a shame.

When we developed Familiar, we tried to use other engines for OCR before settling with Apple's native one. the issue is performance. other engines are slower or more resources intensive or both.
Apple's native OCR engine is basically what allows Familiar to be so resource efficient.
Due to that, currently the decision is not to add any other OCR engine into Familiar as a built in option.

Regarding downscaling of the resolution, its another "hack" so that it takes less time to process those images. The results for images with English or Hebrew are pretty good even if the resolution is lower.

ATM I suggest two things:

1. if someone else has issues with that, please comment or add emoji so we have an understanding on how critical this issue is
2. try to replace the current OCR engine with Tesseract on your end and report back on the overall performance and resource utilization