{
  "markdown": "# 🎬 Klaket\n\n**Turn any video into LLM-ready data.**\n\n[![License: AGPL-3.0](https://img.shields.io/badge/license-AGPL--3.0-f5b70f)](LICENSE)\n[![PRs welcome](https://img.shields.io/badge/PRs-welcome-4ade80)](CONTRIBUTING.md)\n[![Self-host](https://img.shields.io/badge/self--host-docker%20compose%20up-ede8e0)](#quick-start)\n\n![Klaket demo](assets/demo.gif)\n\n> A *klaket* is a clapperboard — the tool that syncs sound and image on a film set. **Klaket syncs video with LLMs.**\n\nLLMs read text. The web became readable with scrapers — but video, the largest store of human knowledge, is still locked away. Klaket unlocks it: give it a video URL or file, get back structured, timestamped, LLM-ready data.\n\n```bash\npip install klaket\nklaket ingest \"https://youtube.com/watch?v=...\" --wait\n```\n\n```jsonc\n{\n  \"transcript\": [\n    { \"start\": 14.32, \"end\": 19.80, \"speaker\": \"S1\", \"text\": \"So let's deploy this with docker compose...\" }\n  ],\n  \"scenes\": [\n    { \"start\": 190.0, \"end\": 342.5, \"keyframes\": [\"scene_004_01.jpg\"] }\n  ],\n  \"chapters\": [...],\n  \"summary\": \"...\"\n}\n```\n\n## Features\n\n- **📝 Transcript** — timestamped speech-to-text in **~100 languages** (auto-detected) with **word-level timestamps**; pick the model per job (`\"model\": \"medium\"`)\n- **🎙️ Podcasts too** — pass an audio file/URL (mp3, m4a…) and Klaket skips the visual stages, deriving chapters from speech pauses\n- **🗣️ Speaker diarization** — who said what (S1/S2/…), local & keyless (sherpa-onnx)\n- **💬 Subtitles** — ready-to-use `.srt` / `.vtt` files with speaker labels\n- **🎞️ Scene detection** — content-aware scene boundaries + keyframes per scene\n- **🔎 On-screen text (OCR)** — reads slides, terminals and captions per scene, local & keyless\n- **🧩 One JSON timeline** — transcript, scenes, frames and on-screen text aligned on a single timeline\n- **🔌 Works offline, no API key required** — the core pipeline uses zero LLM calls\n- **🧠 Pluggable model layer** — optional scene descriptions via local VLMs (Ollama) or any OpenAI-compatible endpoint (`KLAKET_VLM=off` by default)\n- **🤖 MCP server** — let coding agents \"watch\" any video and find moments inside it\n- **🔍 In-video search** — `GET /v1/jobs/{id}/search?q=…` finds the exact moment\n- **▶️ Playground** — the dashboard plays the video with a click-to-seek, live-highlighted transcript\n\n## SDKs\n\n```python\n# pip install klaket\nfrom klaket import Klaket\nresult = Klaket().process(\"https://youtube.com/watch?v=...\", num_speakers=2)\n```\n\n```ts\n// npm i klaket-sdk\nimport { Klaket } from \"klaket-sdk\";\nconst result = await new Klaket().process(\"https://youtube.com/watch?v=...\");\n```\n\n## Give your agent eyes\n\n```bash\n# Claude Code\nclaude mcp add klaket -- npx klaket-mcp   # KLAKET_API_URL defaults to localhost:8484\n```\n\nThen: *\"Watch https://youtube.com/watch?v=… and summarize the commands the presenter runs.\"*\nThe agent gets `klaket_ingest`, `klaket_job_status` and `klaket_get_result` tools.\n\n## Quick start\n\n```bash\ngit clone https://github.com/huseyinstif/klaket.git && cd klaket\ndocker compose up --build\n# API on :8484, dashboard on :5180\ncurl -X POST localhost:8484/v1/ingest \\\n  -H \"Content-Type: application/json\" \\\n  -d '{\"url\": \"https://youtube.com/watch?v=...\"}'\n```\n\nThat's it — no API keys, no GPUs required. `make help` lists developer shortcuts (`make up`, `make test`, `make e2e`).\n\n## Architecture\n\n```\nclient ──► Go API ──► Redis queue ──► Python worker (ffmpeg · faster-whisper · scenedetect)\n                │                          │\n            dashboard ◄────────────────────┘   /data/jobs/<id>/result.json\n```\n\n- `apps/api` — Go, job orchestration\n- `apps/worker` — Python, media pipeline\n- `apps/dashboard` — React dashboard\n\n## Self-host vs Cloud\n\nKlaket is open source (AGPL-3.0) and fully self-hostable. A hosted, pay-per-minute cloud API with managed GPUs is planned — join the waitlist (coming soon).\n\n## Status\n\n🚧 v0.7 — pre-1.0, moving fast. Star the repo to follow along.\n\n## License\n\n[AGPL-3.0](LICENSE). SDKs and clients will be MIT.\n\n## Contact\n\nBuilt by Hüseyin Tıntaş — [X (@1337stif)](https://x.com/1337stif) · [LinkedIn](https://www.linkedin.com/in/huseyintintas/)\n",
  "bytes": 4149,
  "sha": "19e228fcfe0fec1347b63cf789c0b554c04b82d8bfa5d633cffa80ac1fe16d0a",
  "repo_slug": "huseyinstif/klaket",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_io_github_huseyinstif_klaket_mcp_f5824aa7/readme"
}