Skip to content

Klaket v0.7.0 — first public release

Choose a tag to compare

View all tags
@huseyinstif huseyinstif released this 10 Jul 13:21
v0.7.0
cf9dec5

The first public release of Klaket — turn any video into LLM-ready data.

Highlights

  • Transcripts in ~100 languages with word-level timestamps (faster-whisper), per-job model selection and context prompts
  • Speaker diarization — local & keyless (sherpa-onnx), with word-level splitting and blip smoothing
  • Scene detection + keyframes (PySceneDetect AdaptiveDetector) and on-screen text OCR (RapidOCR)
  • Auto chapters, SRT/VTT subtitles and local subtitle translation (Argos) — no external APIs
  • In-video search: diacritic-insensitive, prefix-aware, typo-tolerant
  • Playlist ingest, batch mode, webhooks, audio-only (podcast) support
  • MCP server so coding agents can "watch" videos, plus Python/JS SDKs and LangChain/LlamaIndex loaders
  • React dashboard with a click-to-seek playground
  • One-command self-host: docker compose up --build — no API keys, no GPUs required

Quality gates

  • 20-check end-to-end suite (make e2e), Go + Python unit tests in CI
  • Battle-tested against real YouTube videos (AV1 normalization, hallucination guards, WER-measured accuracy)
Assets 2
Loading