Semantic Search Engine
Under maintenancePricing
Pay per usage
Go to Apify Store
Semantic Search Engine
Under maintenanceSemantic search engine using embeddings for intelligent document retrieval and similarity search
Build a vector index from your documents and search it semantically. Powered by turbovec — 8-16x compression over float32, faster than FAISS.
Features
- 8-16x memory compression — 10M docs fit in 4GB RAM (vs 31GB as float32)
- No training phase — add vectors, they're indexed immediately
- Local or API embeddings — sentence-transformers or OpenAI
- Filtered search — restrict results to a subset of documents
- Persistent index — save/load from disk, incremental sync
- Framework ready — drop-in for LangChain, LlamaIndex, Haystack
How It Works
- Build mode: Provide documents → embeddings are generated → indexed with turbovec → saved
- Search mode: Provide queries → embedded → searched against index → ranked results returned
Input
| Field | Type | Description |
|---|---|---|
| mode | String | "build" or "search" |
| documents | Array | Texts or {text, id, vector} objects |
| queries | Array | Search queries (search mode) |
| k | Integer | Results per query (max 100) |
| bitWidth | Integer | 2 (16x) or 4 (8x) compression |
| embeddingMode | String | auto, local, api, precomputed |
| embeddingModel | String | Sentence-transformers model name |
Output
Build mode returns index metadata (vector count, dimension, compression ratio). Search mode returns ranked results with scores for each query.
Use Cases
- Semantic search on scraped web data
- RAG pipeline for AI agents
- Document retrieval for LLM context
- Privacy-preserving local search (no data leaves your machine)
Environment Variables
OPENAI_API_KEY— Required for API embedding mode
