← 回產品頁
Source-available AI metadata layer for DIT workflows — Resolve-native, CJK-first.
arkiv attaches AI metadata (transcript, vision tags, atmosphere, edit position) to your footage and surfaces clips via natural-language search in Chinese, Japanese, or English.
Step 1 — Install dependencies (macOS)
brew install python ffmpeg ollama
Linux / Windows: see install.md.
Step 2 — Clone and set up
git clone https://github.com/vulture-s/arkiv.git && cd arkiv
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
pip install mlx-whisper # Apple Silicon
# pip install faster-whisper torch # NVIDIA GPU
Step 3 — Pull Ollama models
ollama pull bge-m3 # required
ollama pull qwen2.5vl:7b # optional: vision tags
ollama pull qwen2.5:14b # optional: transcript polish
Step 4 — Ingest media
python ingest.py --dir /path/to/footage
python embed.py
Step 5 — Search
uvicorn server:app --host 0.0.0.0 --port 8501
# Open http://localhost:8501
python health.py # all required checks should PASS
Source-available under the PolyForm Perimeter License 1.0.1 — arkiv is free to use for any purpose, commercial work included, and what you make with it is yours without restriction; the one prohibited use is providing others with a competing product. Self-hosted, no cloud, no telemetry.