Examples

Practical examples demonstrating Cortex memory primitives in real-world scenarios.

RAG Pipeline

Build a retrieval-augmented generation system using the knowledge store.

Agent Memory

Implement persistent conversation memory for long-running agents.

Workflow State

Use context storage for cross-step state in multi-stage workflows.

Entity Tracking

Track people and organizations across interactions.

Quick Examples

Knowledge-Augmented Chat

# 1. Ingest your documentation
cortex knowledge ingest-dir \
  --collection docs \
  --dir ./documentation \
  --pattern "*.md"

# 2. Search for relevant context
cortex knowledge search "authentication setup" --mode hybrid --limit 3

# 3. Use results in your LLM prompt
# (Results provide grounded context for responses)

Persistent Agent Memory

# Agent stores conversation as it runs
cortex conversation append \
  --thread-id agent-session-123 \
  --role assistant \
  --content "I found 3 relevant documents about authentication."

# Later, retrieve context
cortex conversation history --thread-id agent-session-123 --limit 20

Cross-Run State

# Store state that persists
cortex context set "project/analyzed-files" '["main.go", "handler.go"]'

# Next run retrieves it
cortex context get "project/analyzed-files"

Entity Relationships

# List people mentioned across conversations
cortex entity list --type person --sort mention_count

# See who works where
cortex entity relationships ent-jane-123 --type works_at

Integration Patterns

With Iris SDK for Embeddings

Cortex uses the Iris SDK directly—no separate Iris service required. Configure your provider:

embedding:
  provider: openai          # Or: anthropic, voyageai, gemini, ollama
  model: text-embedding-3-small
  dimensions: 1536
  batch_size: 100
  cache_size: 1000

Set your API key:

# For OpenAI (most common)
export OPENAI_API_KEY="sk-..."

# For local inference with Ollama (no API key needed)
export OLLAMA_BASE_URL="http://localhost:11434"

The SDK handles batching, caching, and error recovery automatically.

With PetalFlow for Orchestration

Use Cortex as memory backend for PetalFlow workflows:

// In a PetalFlow node, store state
ctx.ContextSet("user-preferences", prefs)

// Later nodes retrieve it
prefs, _ := ctx.ContextGet("user-preferences")

// Search knowledge for grounding
docs, _ := ctx.KnowledgeSearch("billing questions", 5)

With MCP Clients

Any MCP client can use Cortex tools:

{
  "mcpServers": {
    "cortex": {
      "command": "cortex",
      "args": ["serve"]
    }
  }
}