screenlens.stub

---

id: screenlens name: ScreenLens repo_path: /sessions/charming-sleepy-ritchie/repos/screenlens remote_url: https://github.com/sw30labs/screenlens.git primary_language: Python framework: LangGraph + mlx-vlm + OpenCLIP + ChromaDB ingested: 2026-04-11 last_commit_date: 2026-04-11

ScreenLens

Purpose

Local video scene intelligence for Apple Silicon. LangGraph-orchestrated pipeline: smart keyframe extraction (SSIM + pHash + HSV), Qwen3.5-VL captioning via mlx-vlm, CLIP embedding, ChromaDB vector search, Ollama summarization.

Architecture

LangGraph StateGraph pipeline with two paths:

• Ingest: Extract keyframes → Caption → Embed → Store
• Search: Query → Semantic search → Summarize

Hybrid keyframe detection:

• SSIM (structural similarity) for scene changes
• pHash (perceptual hash) for visual similarity
• HSV histogram for color distribution

Key Technologies

• Framework: LangGraph, MLX
• Vision-Language: Qwen3.5-VL 122B via mlx-vlm
• Embedding: OpenCLIP
• Vector DB: ChromaDB
• CLI: Typer + Rich

Dependencies

langgraph, langchain, langchain-ollama, langchain-community, chromadb, open-clip-torch, opencv-python-headless, scikit-image, Pillow, numpy, torch, tqdm, typer, rich, pydantic, mlx-vlm, huggingface-hub

Tags

python, langgraph, mlx, apple-silicon, local-inference, chromadb, video, ocr

Cross-references

• qwen3-vl-python-ocr-script-mlx — shared Qwen3 VL models