Features
Hybrid search combining dense vectors, sparse vectors, keywords, filters, and custom scoring
Multi-vector retrieval with late interaction (SMVE)
Vector search (dense and sparse)
Keyword search with BM25 scoring
File Search with grounded answers and citations
Unified retrieval engine for structured and unstructured data
Object storage-based storage for lower cost and unlimited scale
Sub-100ms latency at billion-scale
Native SDKs for Python, JavaScript, Rust, and SQL compatibility (Postgres wire protocol)
Custom scoring to combine multiple ranking signals
Document processing (PDFs, DOCX, images, text) with OCR, parsing, and embedding
Built-in multi-vector embedding for semantic search
Agentic and research query support for File Search
Private deployment (VPC or on-premises)
Web crawling and scraping API with Rust engine
AI Studio for natural language crawling (add-on $6/mo)
Browser AI commands via WebSocket: Act, Extract, Observe
Silk custom AI model for extraction and captcha solving
Browser Cloud with stealth anti-detection
Structured output: markdown, HTML, JSON, CSV, XML, plain text
Screenshot capture of pages
Link extraction from pages
Search endpoint for query-based data retrieval
Unblocker with rotating proxies and automatic retries
1,000+ ready-made scraper examples (32 categories)
Data connectors: S3, GCS, Google Sheets, Azure Blob, Supabase
Respects robots.txt (configurable)
Failed requests not billed
Open-source core available on GitHub
