Cortex Search

What it does

Cortex Search is a retrieval service that takes text columns from a Snowflake table or view and builds a hybrid vector and keyword search index over them. You define a search column, optional attribute columns for filtering and returned results, and a target lag that controls how often the index refreshes from the base data. The service then returns ranked matches for a query, which can feed a chatbot or a search bar in an application.

People typically use it as the retrieval layer for retrieval augmented generation, so a large language model answers from current proprietary data, or as the backend for an enterprise search box. Creation and refresh run on a virtual warehouse you specify, and embeddings are computed automatically for inserted or changed rows, so teams avoid managing their own embedding pipeline or index tuning.

It suits data and application teams already working in Snowflake who have text-heavy content such as support transcripts or documents. Access requires a role granted the SNOWFLAKE.CORTEX_USER database role, plus a warehouse and usage costs for embedding, serving, storage, and warehouse compute. Serving charges continue while the service is resumed, and some features such as stage sources are labeled Preview.

What it can do

• Builds a hybrid vector and keyword search index over text columns in a Snowflake table or view
• Computes text embeddings automatically for inserted or updated rows
• Supports attribute columns for filtering and returning fields alongside search results
• Refreshes the index on a configurable target lag tied to base-data changes
• Provides a SEARCH_PREVIEW function to check service contents from SQL
• Offers a Python API to retrieve the most relevant records for a query

Works with

• Cortex Agents (uses Cortex Search as the retrieval layer)
• Cortex LLM Functions (combined for RAG chatbots)

Before you choose

• Serving compute is billed per GB per month of indexed data while the service is resumed, even without queries; suspend or use AUTO_SUSPEND when idle.
• Changing the source query schema or using CREATE OR REPLACE on the base table triggers a full re-embedding; incremental updates such as MERGE INTO are cheaper.
• Embedding cost applies per token for each inserted or updated row, so estimate tokens with COUNT_TOKENS on a representative sample.

Listed under

• Data

Get started

Requires a role with SNOWFLAKE.CORTEX_USER and a warehouse.

1. Create a database, schema, and dedicated X-SMALL or appropriately sized warehouse (MEDIUM or smaller recommended) for the service.
2. Grant a role the SNOWFLAKE.CORTEX_USER database role, then create the service with CREATE CORTEX SEARCH SERVICE or via Snowsight under AI & ML » Cortex Search.
3. Preview the service with SNOWFLAKE.CORTEX.SEARCH_PREVIEW and grant USAGE on the database, schema, and service to the roles that need access.

Charges are usage-based and cover embedding the text in new or changed rows, an ongoing serving charge for the indexed data while the service is resumed (even with no queries), storage for the materialized source table and index, virtual warehouse compute for building and refreshing the index, and cloud services compute. Cloud services are billed only when the daily cloud services cost exceeds 10% of the daily warehouse cost. Rates are listed in the Snowflake Service Consumption Table.

Source: docs.snowflake.com/en/user-guide/snowflake-cortex/cortex-search/cortex-search-costs, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Cortex Search | Snowflake Documentation ↗docs.snowflake.com
• Understanding cost for Cortex Search Services | Snowflake Documentation ↗docs.snowflake.com

Company website

docs.snowflake.com: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.