Skip to content
Rime Docs

Cortex AI

Snowflake Cortex is Snowflake’s built-in suite of AI and machine learning services. It runs directly within your Snowflake account, meaning your data never leaves Snowflake’s security boundary for AI processing. Rime integrates with Cortex to let you enable and manage AI capabilities over the data your pipelines produce.

What Snowflake Cortex provides

Cortex includes three main service categories that Rime supports:

Search Services

Cortex Search enables semantic search over text data stored in your Snowflake tables. Unlike traditional SQL LIKE or full-text search, semantic search understands meaning — a query for “customer complaints about delivery” will match records about “late shipping” or “package not received” even if those exact words are not in the query.

Use cases:

  • Searching support tickets or customer feedback stored in your warehouse
  • Finding related documents across large text datasets
  • Enabling natural language lookups against reference data

Cortex Analyst

Cortex Analyst provides natural language querying of your data. Instead of writing SQL, users describe what they want in plain English, and Cortex Analyst generates and executes the appropriate query against your data model.

Use cases:

  • Enabling non-technical users to query the data warehouse directly
  • Ad-hoc data exploration without writing SQL
  • Quick answers to business questions during meetings or reviews

Cortex Agents

Cortex Agents are autonomous AI agents that can query, analyze, and reason about your data. An agent can execute multi-step workflows — querying multiple tables, performing calculations, and synthesizing results into a coherent answer.

Use cases:

  • Automated data quality analysis (“Are there any anomalies in this week’s sales data?”)
  • Root cause investigation (“Why did revenue drop in the Auckland region last month?”)
  • Report generation from natural language prompts

Prerequisites

Before enabling Cortex AI through Rime, ensure:

  1. Snowflake Enterprise Edition (or higher). Cortex services are not available on Standard Edition. Check your Snowflake account edition in the Snowflake console under Account > Usage.

  2. Cortex enabled in your Snowflake account. Cortex must be activated by your Snowflake account administrator. This is a Snowflake-side setting, not something Rime controls.

  3. Sufficient Snowflake credits. Cortex services consume Snowflake credits based on usage. Ensure your credit quota accounts for AI workloads.

  4. Appropriate permissions. The Snowflake role used by Rime needs USAGE privileges on the Cortex functions and the tables/views you want to make searchable.

Setup in Rime

Connecting to Cortex services

Navigate to Project > Settings > Cortex AI to configure the integration:

  1. Verify connection. Rime checks that your Snowflake connection has access to Cortex services. If Cortex is not available, you will see a message explaining what is missing.

  2. Select a warehouse. Choose which Snowflake warehouse Cortex queries should run on. We recommend a dedicated warehouse for AI workloads to avoid contention with your regular pipeline queries. Size the warehouse based on expected query volume.

  3. Enable services. Toggle on the Cortex services you want to use: Search, Analyst, Agents, or any combination.

Configuring searchable datasets

For Cortex Search, you need to specify which tables or views contain the text data you want to search:

  1. Go to Cortex AI > Search Configuration
  2. Click Add Dataset
  3. Select the database, schema, and table
  4. Choose the text columns to index for search
  5. Optionally add metadata columns that should be returned alongside search results (e.g., timestamps, IDs, categories)

Rime creates the necessary Cortex Search Service objects in Snowflake and keeps them updated as your data changes. When a connector sync or dbt run updates the underlying tables, the search index is refreshed automatically.

Configuring Cortex Analyst

Cortex Analyst works against a semantic model that describes your data in business terms. Rime generates this semantic model from your transformation projects:

  1. Go to Cortex AI > Analyst Configuration
  2. Select the transformation project (typically your mart layer) that represents your business data model
  3. Review the generated semantic model — Rime maps table and column names, data types, and relationships into a format Cortex Analyst understands
  4. Optionally add business descriptions to tables and columns to improve query accuracy (e.g., describing what “ARR” means or what the “region” column contains)

The semantic model is regenerated automatically when your transformation project changes.

Configuring Cortex Agents

Agents are configured with a scope (which tables and functions they can access) and optional instructions:

  1. Go to Cortex AI > Agent Configuration
  2. Click Create Agent
  3. Give the agent a name and description (e.g., “Sales Analyst”, “Data Quality Checker”)
  4. Select the tables and views the agent can query
  5. Optionally provide instructions that guide the agent’s behavior (e.g., “Always express monetary values in NZD”, “When asked about growth, compare to the same period last year”)

Each agent runs with the permissions of the Snowflake role configured in your Rime connection. Agents cannot access tables outside the scope you define.

Usage monitoring

Cortex AI usage is tracked in Project > Settings > Cortex AI > Usage:

MetricDescription
Search queriesNumber of semantic search queries executed
Analyst queriesNumber of natural language queries processed
Agent invocationsNumber of agent tasks executed
Credit consumptionSnowflake credits consumed by Cortex services
Average latencyMean response time for each service type

Usage data is broken down by day and by service type. You can set up alert rules to monitor Cortex credit consumption and get notified if usage exceeds a threshold.

Cost tracking

Cortex services consume Snowflake credits. The cost depends on the warehouse size, query complexity, and volume of requests. Rime tracks credit consumption attributed to Cortex workloads separately from your pipeline credits, so you can see exactly how much AI features cost.

Cost tracking is available in the usage panel and in the DORA metrics dashboard (under the infrastructure cost section if enabled).

To control costs:

  • Use a smaller dedicated warehouse for Cortex queries
  • Set auto-suspend on the Cortex warehouse to avoid idle credit consumption
  • Configure credit alerts through Rime’s monitoring to catch runaway usage

Permissions and security

Cortex AI respects Snowflake’s existing role-based access control:

  • Cortex Analyst and Agents can only query tables that the configured Snowflake role has SELECT privileges on
  • Masking policies applied through Rime are enforced when Cortex queries the data — masked columns return masked values even through AI queries
  • Search indices inherit the access controls of their source tables

This means your existing governance configuration applies automatically to AI features without additional setup.

Limitations

  • Cortex services are only available in certain Snowflake regions. Check Snowflake’s documentation for region availability.
  • Cortex Analyst accuracy depends on the quality of your semantic model. Tables and columns with clear, descriptive names produce better results than abbreviated or cryptic names.
  • Agent capabilities are determined by Snowflake’s Cortex release. New features appear as Snowflake adds them.
  • Large text datasets may take time to index for search. The initial indexing runs in the background and does not block other operations.

Next steps