Skip to main content

Quickstart

Get a Cortex agent running locally in four steps.

1. Install Cortex​

pip install cortex-runtime==2.5.10

2. Initialize and Build​

Unlike legacy versions, Cortex v2.5.10 uses a build-first workflow to ensure environment consistency and optimize the Shared KV Cache.

# Detect architecture and prime the local cache
cortex build

3. Define your agent​

Create a cortex.yaml in your project directory:

version: "2"
project: my-first-agent

agents:
  - name: assistant
    runtime: python3.12
    entrypoint: agent:run
    resources:
      cpu_limit: 1.0
      mem_limit: 1Gi
    capabilities:
      - network.access: ["api.openai.com"]
      - turbo_mode: true # Enable 3x parallel speedup

4. Write the agent logic​

Create agent.py in the same directory:

import os
from openai import OpenAI

def run():
    client = OpenAI(base_url=os.environ.get("ICE_URL", "https://api.openai.com/v1"))

    while True:
        user_input = input("You: ")
        if user_input.lower() in ("exit", "quit"):
            break

        response = client.chat.completions.create(
            model="gpt-4o",
            messages=[{"role": "user", "content": user_input}],
            extra_headers={
                "X-Session-Id": os.environ.get("CORTEX_SESSION_ID", "default"),
                "X-User-Id": os.environ.get("CORTEX_USER_ID", "user")
            }
        )
        print(f"Agent: {response.choices[0].message.content}")

5. Run it​

# Bundle the project
cortex bundle --config cortex.yaml --output my-first-agent.cortex

# Run it
cortex run my-first-agent.cortex

Or run directly without bundling during development:

cortex dev --config cortex.yaml

Next Steps​