Skip to content
Cortex Docs

Embeddings & Vectors

import { Tabs, TabItem, Aside } from ‘@astrojs/starlight/components’;

Cortex uses vector embeddings to power its Semantic Search and Intelligent Context Routing features.

Local-First with Ollama

By default, Cortex attempts to use Ollama for local embeddings. This ensures your data never leaves your machine.

Requirements

  • Ollama installed and running.
  • The nomic-embed-text model pulled:
    Terminal window
    ollama pull nomic-embed-text

Configuration

Cortex automatically detects Ollama if it’s running on http://localhost:11434.

Cloud Fallback with OpenAI

If local embeddings are unavailable or too slow, you can use OpenAI.

Configuration

Set the OPENAI_API_KEY environment variable in your shell or .env file.

Terminal window
export OPENAI_API_KEY="sk-..."

Advanced: SQLite-Vec

Cortex is moving towards native, high-performance vector storage using sqlite-vec.

Current Status

In v0.8.0, semantic search uses a hybrid approach:

  1. Keyword search via FTS5 for precision.
  2. In-memory cosine similarity for semantic relevance.