TopK is a paid tool. Please visit their website for pricing details.

TopK

A managed hybrid retrieval engine for RAG, agent context, and recommendations, billed monthly by usage.

subscription

Last Updated 8/28/2026

About TopK

TopK is a managed hybrid retrieval engine for AI applications. Its single API covers dense and sparse vector search, BM25, multi-vector retrieval, attribute filters, custom scoring, managed inference, and document processing. It handles structured data as well as private files; the datasets workflow can ingest documents, search them, and return cited answers. Developers can connect through Python, JavaScript, Rust, SQL, a CLI, or an MCP server. Durable data sits in object storage, separate from the compute that serves queries. Billing is usage based and monthly, with separate charges for storage, writes, query requests, memory, scanned data, and egress. The $10 starting credit makes a small evaluation possible, but a production budget needs realistic workload estimates. TopK fits teams whose search quality requires more than nearest-neighbor vector matching, especially for RAG, agent context, or recommendation candidate generation.

Best for

Search and AI engineering teams that need hybrid retrieval and custom scoring in one managed API.

RAG builders working with private files who need parsing, OCR, retrieval, answers, and citations in one workflow.

Recommendation teams that need fresh writes, filters, and several representation types in a single serving layer.

Best use cases

Use TopK when a RAG pipeline must search PDFs or scans and return answers linked to the exact source.

Use TopK when candidate generation combines vector similarity, keywords, attribute filters, and application-specific scoring.

Use TopK when agents need private context through an API, command-line workflow, or MCP connection.

Compare before choosing

Compared with a basic vector database, TopK supports BM25, sparse and multi-vector retrieval, filters, and custom scoring in one query.

Compared with stitching together parsing, embeddings, retrieval, and reranking services, TopK places these functions behind one platform API.

Compared with fixed-seat pricing, metered billing follows storage and query behavior but requires workload estimates to forecast cost.

Features

RAG document search

Ingest PDFs, scans, manuals, policies, and filings, then answer questions with citations tied to the source material.

Recommendation retrieval

Generate candidates from dense, sparse, keyword, and multi-vector representations while applying fresh writes and product-aware filters.

Agent context search

Connect private data to agents through the API, CLI, or MCP server and retrieve relevant context with hybrid scoring.

Pros

One API combines vector, keyword, filtering, multi-vector, and custom-scoring workflows.
Document workflows include parsing, OCR, retrieval, answer generation, and source citations.
Python, JavaScript, Rust, SQL, CLI, and MCP access cover several application environments.

Cons

Metered billing spans several dimensions, so forecasting requires realistic workload estimates.
Query memory is priced separately at $20 per GiB, which can materially affect serving cost.
Adoption requires a TopK account and API key, adding a managed external service dependency.

Pricing Plans

Usage-based

MeteredMonthly

Storage: $0.10 per GiB
Write requests: $0.01 per 1K writes
Write data: $0.50 per GiB written
Query requests: $0.50 per 1M queries
Query memory: $20.00 per GiB
Query compute: $0.005 per TiB scanned
Data egress: $0.20 per GiB returned
$10 in starting credits

* Pricing information is for reference only. Please refer to the latest announcement on TopK's official website.

FAQ

TopK is a managed search engine for AI applications, combining semantic and keyword retrieval, filters, multi-vector search, and custom scoring in one query.

TopK runs dense and sparse vectors, BM25, attribute filters, late-interaction retrieval, and custom scoring against one index instead of limiting retrieval to one embedding per record.

TopK bills monthly for measured storage, writes, queries, query memory, scanned data, and data returned. New accounts can begin with $10 in free credits.