Skip to content

CapsuleCraft v1.12.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 02 Nov 06:09
· 22 commits to main since this release
v1.12.0
c96b66c
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

CapsuleCraft v1.12.0 — Integrations

Local AI you download in the app, plus native PDF and image viewing.

The headline

The native shell now has an Integrations catalog (⚙ in the title bar):
offline engines that appear in the app only after you explicitly download
them. Nothing is bundled, the vault stays a plain folder of .eml files, and
every component installs into the app-data directory — never into the vault.

  • OCR (Xiaomi-OCR-0) — Xiaomi's new OCR model (Qwen3.5-0.8B, Apache-2.0)
    served by mistral.rs, a pure-Rust engine. No llama.cpp, no Python, no
    cloud. Recognize text on any image attachment with one button; the result
    becomes a paragraph on the page. The engine is built for all three desktop
    platforms in this release's CI and attached as mistralrs-engine-*.tar.gz.
  • STT (whisper) — tiny/base/small run in-process on candle (pure Rust,
    CPU). One button in the audio block transcribes a recording into its
    transcript field; Russian is forced via the language token.
  • TTS (Русские голоса) — Ирина and Дмитрий (piper voices) through the
    sherpa-onnx runtime; the 🔊 button in the document editor speaks the page
    through the existing audio player.
  • PDF viewer — page navigation and zoom, rendered natively via pdfium.
  • Image viewer — PNG/JPEG/GIF/WebP with zoom.

Notes

  • The largest component is the OCR model (~1.75 GB); the rest is 4–70 MB each.
  • pdfium and sherpa-onnx are renderer/codec runtimes loaded at runtime — the
    only non-Rust pieces in the feature. Inference itself is Rust end to end.
  • Downloads do not resume; a failed download can simply be retried from the
    catalog card.
Assets 15
Loading