Frequently
asked questions.

TopK is a managed search engine for AI applications. It lets you build search, RAG, and agent context systems using semantic search, BM25, sparse vectors, multi-vector retrieval, metadata filters, and custom hybrid scoring in one query.

Instead of stitching together a vector database, keyword index, embedding pipeline, reranker, and application-side ranking logic, TopK gives you one engine for high-quality retrieval at production speed.

Most vector databases are optimized around nearest-neighbor search over one embedding per chunk or record. That works for simple semantic search, but production-grade search is usually more complex.

TopK is a search engine: dense & sparse vectors, late interaction, BM25, metadata filters and custom scoring run in one query against a single index. This lets you optimize retrieval natively inside the engine without having to build and manage retrieval logic in your application code.

Single-vector retrieval compresses a document, passage, image, code file, or table into one embedding. This compression is lossy and can miss important details, especially in long documents or queries with multiple intents.

Multi-vector retrieval stores richer representations, such as token-level or patch-level vectors. At query time, TopK can match each part of the query against the most relevant part of the document, which helps with long documents, code identifiers, rare terms, tables, charts, policies, and multi-part questions.

Yes, TopK supports custom dense, sparse, and multi-vector embeddings out of the box.

Sometimes, but not as the foundation of your retrieval system.

Rerankers are useful as a final precision layer, especially when you want an expensive model to review a smaller candidate set. But many teams use rerankers to compensate for weak retrieval. TopK helps improve first-stage retrieval quality and reduces the need for a dedicated reranker stage. That being said, you can use TopK with or without a reranker, depending on your latency, cost, and quality requirements.

No. File search is one important use case, but TopK is a general-purpose search engine.

We make it easy to ingest unstructured and multi-modal data (for example, PDFs, Markdown, HTML, ...) and get highly accurate answers with citations. This allows you to build agents on top of private knowledge bases with the full flexibility of our search engine.

Yes. We built TopK with least common denominator dependencies (VMs and object storage) to make it easy to deploy anywhere. We support all major cloud providers and on-prem options with full data-residency and network isolation guarantees.

TopK is SOC 2 Type I certified. We encrypt your data at rest and in transit, provide role-based access control, audit logging, and VPC deployment options. For teams with specific regulatory requirements — HIPAA, GDPR, or others — contact us to discuss your compliance needs.

TopK ships with a Python SDK, a TypeScript SDK, a CLI, and a SQL interface so you can connect from backend services, agents, and existing PostgreSQL-compatible tools.