SovLens
Your Media. Your AI. On Your Hardware.
SovLens is a native desktop application that builds a fully private, semantic index of your photos and videos. Find specific frames and images using natural language queries — running entirely on your local GPU or CPU. Zero cloud dependencies, zero subscription fees.
Explore the interactive desktop interface
Test-drive the actual SovLens application interface. Click through the library sections in the sidebar, or select one of the sample search queries to simulate a private semantic vector query.
All Media
Drag & drop files to ingest
Images and videos are processed 100% locally
Total: $14.50
Date: May 3 2026
EasyOCR scan
How local AI indexes your media
SovLens works entirely on your local machine by downloading public pre-trained models. When you select a library folder, the app pipelines your files through four local models:
Visual & Semantic
CLIP (OpenAI)Converts video frames and photos into numerical vector representations. Translates human text queries into the same vector space to find matches instantly without tagging.
Local Voice-to-Text
Whisper (OpenAI)Transcribes audio tracks from video clips locally on your device. Search for spoken phrases to jump to the exact timestamp where the words were spoken.
Image Text OCR
EasyOCRScans all image frames for text. Indexing screenshots, whiteboard drawings, slides, and written documents, making them searchable with keyword queries.
Object Detection
YOLOv8 (Ultralytics)Crops and indexes individual elements inside a picture. Perfect for isolating small items, objects, or logos in larger cluttered photos and videos.
Scale analysis to fit your hardware limits
Medium Analysis (Default)— Optimized for CPU / Apple Silicon MPS
The sweet spot for most personal libraries. Delivers fast index times with great search accuracy across photos and video clips.
Semantic Image Search
Model: CLIP (ViT-B/32) • Sampling: Every 3 seconds
OCR Text Recognition
Disabled
Video Audio Transcription
Disabled
YOLOv8 Object Detection
Disabled
Desktop tech stack built for speed
Tauri V2 Shell
Wraps the frontend in a lightweight, secure native binary. Replaces chromium wrappers like Electron, shrinking bundle sizes to under 80 MB.
FastAPI Backend
Runs a lightweight local python server behind the app. Loads CLIP and Whisper weights on-demand, handling vector calculations with low overhead.
LanceDB Database
Serverless vector database designed for lightning fast search. Keeps similarity searches under 100ms even on libraries containing millions of images.
GPU Optimization
Accelerates execution using MPS on Apple Silicon MacBooks and CUDA on NVIDIA GPUs, automatically falling back to optimized multithreaded CPU loops elsewhere.
Why we built SovLens
Cloud photo indexing works by uploading your memories, receipts, and private screenshots to corporate servers. Once there, they are used to build profiles and train massive models, locked behind monthly subscriptions. SovLens returns to an older, sovereign computational model:
"Your hardware should do the work, and your files should stay on your disk."
By combining open models with rust-based app shells and modern local vector databases, we created a search utility that is just as fast and smart as cloud photo libraries — without the subscription fees or the corporate surveillance.