Quickstart
Get up and running with Cortex in 5 minutes.
This guide walks you through installing Cortex, pushing your first memory, and connecting an AI agent.
Prerequisites
- Node.js 18 or later
- A Cortex instance (cloud or self-hosted)
- An agent key (obtained from the dashboard)
Step 1: Install the CLI
npm install -g @cortex-labs/cli
Verify the installation:
cortex --version
Step 2: Authenticate
Log in with your agent key from the dashboard:
cortex login --key crtx_YOUR_KEY_HERE --url https://your-cortex-host
Agent keys look like crtx_... and are created in the Cortex dashboard under Agents. Each key authenticates as the human it acts for, so every write is scope-filtered and audited.
Verify your identity:
cortex whoami
Expected output:
John Doe <john@company.com> — role IC, level Senior
Endpoint: https://your-cortex-host
Step 3: Push Your First Memory
Create a file with some knowledge you want to store:
# API Rate Limiting
All external API endpoints must implement rate limiting:
- Public APIs: 100 requests/minute per IP
- Authenticated APIs: 1000 requests/minute per user
- Internal APIs: unlimited
Use Redis sliding window for tracking.
Push it to Cortex:
cortex push ./api-conventions.md --title "API Rate Limiting Policy" --scope engineering
Step 4: Connect an AI Agent
Claude Code
claude mcp add --transport http cortex https://your-cortex-host/mcp \
--header "Authorization: Bearer crtx_YOUR_KEY_HERE"
Cursor
Add to .cursor/mcp.json:
{
"mcpServers": {
"cortex": {
"url": "https://your-cortex-host/mcp",
"headers": {
"Authorization": "Bearer crtx_YOUR_KEY_HERE"
}
}
}
}
Step 5: Recall Memories
Your agent can now search Cortex. Try asking Claude:
"What's our rate limiting policy for external APIs?"
The agent will use the recall tool to find relevant memories with full citations.
What Just Happened?
sequenceDiagram
participant User
participant CLI as Cortex CLI
participant API as Cortex API
participant DB as PostgreSQL + Qdrant
User->>CLI: cortex push api-conventions.md
CLI->>API: POST /api/memories
API->>API: Resolve scope & clearance
API->>DB: Insert memory + vector
API-->>CLI: Memory stored with ID
CLI-->>User: ✓ Pushed memory
Note over User,DB: Later...
User->>User: Asks Claude about rate limiting
Claude->>API: recall("rate limiting policy")
API->>API: Check RBAC permissions
API->>DB: Vector search (scoped)
API-->>Claude: Memory with citation
Claude-->>User: Answer with source