Skip to content

v0.1.12

Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 17 Oct 15:21
v0.1.12
486fdbc
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

TextFlowKit v0.1.12

New

  • Optional AI transcript cleanup in the local browser interface, using an explicitly selected, already-installed local Ollama model. Cleanup is off by default; no cleanup model or Ollama installation is bundled.
  • Preserve the original transcript and exports. Save the AI-corrected result as a separate variant, with an optional side-by-side comparison and its own text download.
  • Bounded cleanup requests, strict complete segment coverage, and persistent review warnings: AI corrections can be guesses and should be checked against the audio.
  • Native Windows installer build tooling and packaging recipe, including isolated runtime/dependency handling and startup/installer coordination tests.
  • Updated README, installation guide, user manual, architecture diagrams, changelog, and landing page for v0.1.12.

Important: Windows installer binary is not published

This release publishes source code and Python distributions, not a Windows setup EXE. The locally built installer remains held while we verify and provide the complete corresponding sources for the FFmpeg build and its linked dependencies. Tracked publicly in #31. There is no public installer download in this release. Python packages do not bundle that FFmpeg binary.

Verification scope

  • Merged-main CI passed all 16 workflow jobs: Windows/macOS/Linux across Python 3.10–3.13, installed-wheel checks on all three systems, and Ruff. The exact tagged commit is 486fdbc04a8afe0b29aacb129d78c45a3b44dab7. Platform/provider skips remain skips, not reported passes.
  • Real local cleanup evidence: 39 segments processed in three bounded blocks, with four edits; the original transcript remained unchanged. This is functional proof, not an audio-accuracy certification of AI guesses.
  • Real native Windows YouTube check passed on the exact merged commit 486fdbc04a8afe0b29aacb129d78c45a3b44dab7 at 2026-10-17 15:11:50 UTC: Whisper tiny/CPU returned three timestamped segments containing the expected word "elephants". This is one URL, not every supported site, and is not hosted-CI evidence.
  • Fonts remain independently versioned at 0.1.6; the attached fonts distributions reuse the already-published bytes.

Install

pip install "textflowkit[export,mcp,http,streaming]==0.1.12"

The local browser interface is launched with textflowkit-ui. FFmpeg and the selected transcription engine's requirements remain separate dependencies; see the installation guide.

User manual · Website · Full changelog

Post-publication checks

Passed on native Windows after the public upload:

  • All four PyPI/GitHub distribution pairs match byte-for-byte and match SHA256SUMS.
  • Fresh isolated install with export,mcp,http,streaming; dependency check passed with no Torch installation.
  • All four commands report 0.1.12; doctor passed.
  • Default Whistle speech selftest passed using existing verified assets in offline mode.
  • The installed HTTP API exported a valid multilingual PDF from a synthetic transcript (in-process ASGI check, not a browser/server end-to-end claim).
  • The Cloudflare production landing page displays v0.1.12, the cleanup documentation, current manuals, and the held-EXE notice.
Assets 7
Loading