Snowflake Cortex brings AI directly to the Snowflake platform. Instead of sending data to external LLM providers or managing GPU infrastructure, Cortex runs models inside Snowflake's secure compute environment.

Your application connects to both Postgres (for data) and Snowflake Cortex (for AI) — Cortex handles the heavy AI workloads while Postgres remains your operational database.

Your application maintains two connections:

1. Postgres — Where your data lives. Query it directly with standard SQL.
2. Snowflake Cortex — Where AI runs. Call Cortex functions via a Snowpark session.

The data stays in Postgres. You call Cortex remotely for AI capabilities — generating text, producing embeddings, or translating content — and use the results in your application or write them back to Postgres.

All Cortex functions are called as SQL expressions through a Snowpark session. Your app sends a query to Snowflake, Cortex processes it, and returns the result — no model hosting required on your end.

Use Cortex COMPLETE to build natural-language query interfaces over your Postgres data. The pattern:

1. Your app sends the user's question plus your Postgres schema context to Cortex COMPLETE.
2. Cortex generates a SQL query.
3. Your app executes that SQL against Postgres.
4. Results come back as a DataFrame you can render as tables or charts.

This gives users a conversational interface to explore data without writing SQL themselves.

Use Cortex to generate vector embeddings and store them in Postgres with pgvector — no need to host an embedding model yourself.

The pattern:

1. Send text content to Cortex AI_EMBED via your Snowpark session.
2. Cortex returns a vector embedding.
3. Store the embedding in a vector column in Postgres using pgvector.
4. Query with pgvector's distance operators for similarity search.

Once embeddings are stored in Postgres, you can use pgvector for fast similarity search at application speed — combining vector retrieval with standard relational filters in a single query.

Retrieval-Augmented Generation (RAG) grounds LLM responses in your actual data. With Cortex and Postgres together:

1. Embed your content — Call Cortex AI_EMBED to generate vectors for your documents. Store them in Postgres with pgvector.
2. Embed the user's query — When a user asks a question, call AI_EMBED on their query to get a vector.
3. Retrieve relevant context — Search Postgres with pgvector to find the most similar content.
4. Generate a response — Pass the retrieved context to Cortex COMPLETE to produce a grounded answer.

Everything stays in your application layer — Postgres holds the data and embeddings, Cortex provides the AI capabilities on demand.

For managed semantic search without building your own embedding pipeline, Cortex Search handles chunking, embedding, indexing, and retrieval automatically.

Mirror your Postgres tables to Snowflake with data mirroring, then create a Cortex Search service for instant hybrid search over your operational data.

Snowflake Cortex gives your Postgres applications access to LLMs and embedding models without managing AI infrastructure. Your data stays in Postgres, Cortex handles the AI workloads, and your application orchestrates between them.

Snowflake Docs — Cortex AI Functions

Reference for all Snowflake Cortex AI functions including AI_EMBED, AI_COMPLETE, AI_EXTRACT, and more.

Snowflake Docs — Cortex AI Features

Overview of Snowflake Cortex AI capabilities.

GitHub — pgvector Documentation

Open-source vector similarity search extension for PostgreSQL.

Snowflake Docs — Data Mirroring

Mirror Postgres data to Snowflake for analytics and AI processing.