Open-Source Hybrid Search for Postgres
Documentation •
Website
Retake is the fastest way to unlock keyword + semantic search for your database.
By connecting directly to Postgres, Retake accelerates the development of AI applications that
require searching, matching, or analyzing database documents. Retake also provides simple search
SDKs that snap into any Python or Typescript application.
You don’t need to worry about managing separate vector stores and text search engines, uploading and embedding documents, or reindexing data. Just write search queries and let Retake handle the rest.
To get started, run our Docker Compose file:
git clone git@github.com:getretake/retake.git
cd retake/docker
docker compose up
By default, this will start the Retake engine at http://localhost:8000 with API key retake-test-key.
Usage
Python
Install the SDK
The core API is just a few functions.
from retakesearch import Client, Index, Database, Table, Search client = Client(api_key="retake-test-key", url="http://localhost:8000") database = Database( host="***", user="***", password="***", port=5432 dbname="***" ) columns = ["column1"] table = Table( name="table_name", columns=columns ) index = client.create_index("my_index") # Note: The table must have a primary key index.add_source(database, table) index.vectorize(columns) # Keyword (BM25) search query = Search().query("match", column1="my query") response = index.search(query) # Semantic (vector-based) search query = Search().with_semantic("my_query", columns) response = index.search(query) # Neural (keyword + semantic) search query = Search().with_neural("my_query", columns) response = index.search(query) print(response)
Typescript
Install the SDK
npm install retake-search
The core API is just a few functions.
import { Client, Database, Table, Search } from "retake-search"; import { withSemantic, withNeural, matchQuery } from "retake-search/helpers"; const client = new Client("retake-test-key", "http://localhost:8000"); // Replace with your database credentials const columns = ["column_to_search"]; const database = new Database({ host: "***", user: "***", password: "***", dbName: "***", port: 5432, }); const table = new Table({ table: "table_name", columns: columns, }); const index = client.create_index("table_name"); // Note: The table must have a primary key index.addSource(database, table); index.vectorize(columns); // Keyword (BM25) search const bm25Query = Search().query(matchQuery("column_to_search", "my query")); index.search(bm25Query); // Semantic (vector-based) search const semanticQuery = Search().query(withSemantic("my query", columns)); index.search(semanticQuery); // Neural (keyword + semantic) search const neuralQuery = Search().query(withNeural("my query", columns)); index.search(neuralQuery);
Key Features
Retake leverages logical-replication-based Change-Data-Capture (CDC) to integrate directly with Postgres. As data changes or new data arrives, Retake ensures that the indexed data is kept in sync.
🧠 Intelligent Vector Cache
Whenever data is changed in Postgres, Retake also updates the embedding/vector representation of that data behind the scenes. Vectors are automatically cached for lightning-fast query results with semantic understanding.
Retake provides intuitive search SDKs that drop into any Python or Typescript application (other languages coming soon). The core API is just a few functions.
Retake enables developers to query with the full expressiveness of the OpenSearch DSL (domain-specific language).
Retake is deployable anywhere, from a laptop to a distributed cloud system.
How Retake Works
A detailed overview of Retake’s architecture can be found in our documentation.
Contributing
For more information on how to contribute, please see our Contributing Guide.
License
Retake is licensed under the Apache-2.0 License.
