How vk-ant/sightrag works
SightRAG is a pluggable visual retrieval system: users index images, videos, or camera input, then search by text or reference image and inspect results. Its documented processing stages include detection, optional segmentation and OCR, embedding, tracking, and storage; retrieval supports optional reranking and multimodal understanding, with visualization and timeline features. The graph includes documented capabilities whose source was not excerpted. Some optional wiring is therefore shown only where the README or sampled source supports it.
Main entry points: core.py SightRAG · indexer.py Indexer · retriever.py Retriever · sqlite_store.py Vector and track store
Components 18
Ingestion
- Visual inputs
- SightRAG
sightrag/core.py - Indexer
sightrag/indexer.py - Frame extraction
sightrag/utils/video.py
Models
- Detection
sightrag/detectors/base.py - Segmentation
sightrag/segmentors/base.py - Embedding
sightrag/embedders/base.py - Inference backends
sightrag/backends/base.py - OCR
sightrag/ocr.py
Retrieval
- Vector and track store
sightrag/store/sqlite_store.py - Retriever
sightrag/retriever.py - Reranking
sightrag/reranker.py - Multimodal understanding
sightrag/multimodal.py
Tracking
- Object tracking
sightrag/trackers/bytetrack.py
Interfaces
- Results and timelines
sightrag/visualizer.py - CLI
sightrag/cli.py - REST API
sightrag/api.py
Other components
- User
Connections 24
- User → SightRAG: uses
- Visual inputs → SightRAG: indexes
- SightRAG → Indexer: dispatches indexing
sightrag/core.py - Indexer → Frame extraction: extracts frames
sightrag/indexer.py - Indexer → Detection: detects objects
- Indexer → Segmentation: segments inputs
- Indexer → Embedding: embeds regions
- Indexer → OCR: reads text
- Indexer → Object tracking: tracks detections
- SightRAG → Inference backends: selects backend
sightrag/core.py - Indexer → Vector and track store: writes vectors
- Indexer → Vector and track store: saves tracks
- SightRAG → Retriever: dispatches queries
sightrag/core.py - Retriever → Embedding: embeds query
sightrag/retriever.py - Retriever → Detection: detects reference
sightrag/retriever.py - Retriever → Vector and track store: searches vectors
sightrag/retriever.py - SightRAG → Reranking: reranks results
sightrag/core.py - SightRAG → Multimodal understanding: understands queries
sightrag/core.py - SightRAG → Vector and track store: reads track history
sightrag/core.py - SightRAG → Results and timelines: renders results
sightrag/core.py - User → CLI: runs commands
- User → REST API: sends requests
- CLI → SightRAG: invokes features
- REST API → SightRAG: serves operations