TopK Documentation

TopK is a hybrid retrieval engine built on object storage for 10x lower cost and massive scale. Semantic, keyword, dense, sparse, and multi-vector search with filtering and custom ranking, in one API.

Set up with a coding agent

Paste this into Claude Code, Cursor, Codex, or any agent with terminal access. It sets everything up from the CLI.

Set up TopK: docs.topk.io/start

Get started

Python SDK

Search from Python. TopK embeds your text for you.

# pip install topk-sdk
client.collection("books").query(
  select(
    score=fn.semantic_similarity(
      "content", "meaning of life"
    ),
  )
  .sort(field("score"), asc=False)
  .limit(10)
)

JavaScript SDK

Search from TypeScript or JavaScript. Same API.

// npm install topk-js
await client.collection("books").query(
  select({
    score: fn.semanticSimilarity(
      "content", "meaning of life"
    ),
  })
  .sort(field("score"), false)
  .limit(10)
);

CLI

Import data and run SQL from your terminal.

brew tap topk-io/topk
brew install topk
topk login
topk project select
topk import ./books.parquet \
  --region aws-us-east-1-elastica

SQL

Search with any PostgreSQL client.

SELECT title,
  semantic_similarity(content,
    'meaning of life') AS score
FROM books
ORDER BY score DESC LIMIT 10;

Search types

Semantic search

Search by meaning. No embedding pipeline, vector store, or reranker to run.

True hybrid search

Blend meaning, keywords, and filters in one query.

Keyword search

Full-text search ranked with BM25.

Dense vector search

ANN search with >95% recall and p99 under 50 ms.

Sparse vector search

Exact retrieval over learned sparse models like SPLADE.

Multi-vector search

Late interaction with MaxSim for long documents.