Skip to content

README_EN.md

Latest commit

 

History

History
120 lines (88 loc) · 3.72 KB

README_EN.md

File metadata and controls

120 lines (88 loc) · 3.72 KB
Raw

vid2note icon

vid2note

Turn video URLs or local media into structured Markdown notes and mind maps.

中文说明 · MIT License · Contributing

vid2note is a local, single-user application without authentication. Docker Compose binds it to 127.0.0.1:8761 by default. Do not expose it to the public internet without authentication, TLS, and network access controls.

Highlights

download → extract audio → ASR → LLM notes → mind map → cleanup
  • YouTube, Bilibili, generic HTTP(S), local video, and local audio input.
  • Experimental bcut online ASR, local whisper.cpp/faster-whisper, or an external ASR endpoint.
  • DeepSeek, Qwen, GLM, Moonshot, MiniMax, Doubao, Baidu, Ollama, and one custom OpenAI-compatible endpoint.
  • Tabbed Settings and ASR diagnostics, including per-provider connectivity tests.
  • Four note-detail levels: concise, balanced, detailed, and exhaustive.
  • Optional PDF handout reference and video-frame embedding.
  • Markdown, XMind, PNG, and Markdown-outline exports.
  • Persistent SQLite tasks, SSE progress, node-level reruns, batch export, and retention rules.
  • One same-origin Docker container for the Vue frontend and FastAPI backend.

Quick start

git clone https://github.com/Goldloli/vid2note.git
cd vid2note
cp .env.example .env
docker compose up -d --build

Open http://localhost:8761 and configure an LLM in Settings. You can also provide the initial configuration through .env:

LLM_PROVIDER=deepseek
DEEPSEEK_API_KEY=your-key
DEEPSEEK_MODEL=deepseek-v4-flash

Public settings stored in data/config/settings.json take precedence over environment defaults.

Ollama

Run Ollama on the host:

ollama pull qwen3.5

Then use:

LLM_PROVIDER=ollama
OLLAMA_MODEL=qwen3.5
OLLAMA_BASE_URL=http://host.docker.internal:11434/v1

Ollama does not need an API key. The Compose configuration provides the host.docker.internal mapping on Linux.

Local ASR

Models are never downloaded automatically. Mount a faster-whisper model directory or whisper.cpp model into the container and set its in-container path on the ASR page. See configuration.

Development

python3.11 -m venv backend/.venv
backend/.venv/bin/python -m pip install -r backend/requirements-dev.txt
cd backend && .venv/bin/python -m pytest

# Node.js 22+
cd ../frontend
npm ci
npm run check

Browser E2E checks use a separate dependency boundary:

backend/.venv/bin/python -m pip install -r backend/requirements-e2e.txt
backend/.venv/bin/python -m playwright install chromium
backend/.venv/bin/python backend/tests/test_playwright.py

Before a public release, scan the tracked tree and reachable Git history. The preflight reports only non-reversible fingerprints, never secret values:

python3 scripts/release_preflight.py

See architecture, troubleshooting, the roadmap, and the security policy.

Privacy and legal notice

Credentials stored through the UI are encrypted with Fernet in data/config/credentials.enc; public settings live in settings.json. Protect and back up credentials.enc and its master key together. For managed deployments, mount the key through VID2NOTE_MASTER_KEY_FILE.

bcut is an experimental online compatibility feature that depends on an external service and may stop working without notice. Video-platform availability is likewise not guaranteed. Only process content you are authorized to use and follow applicable platform terms and law.

Licensed under the MIT License.