Skip to main content
Available on the Core plan and higher. Codex Agent runs on AWS and GCP runners. Contact sales to learn more.
Codex Agent runs the open-source Codex harness in an Ona environment with your project context, environment isolation, integrations, and guardrails.

​
How Codex Agent works

Codex Agent can read your repository, use configured tools, run commands, and open pull requests while following your organization’s policies. On an AWS or GCP runner, Codex Agent uses OpenAI model access through an OpenAI-compatible LLM integration configured for that runner. See Model access for the available options.

​
Prerequisites

  • Your organization is on a Core or Enterprise plan.
  • Codex Agent is enabled for your organization.
  • AWS and GCP runners require OpenAI model access through an OpenAI-compatible LLM integration.

​
Start Codex Agent

  1. Start an environment.
  2. Open the conversation menu.
  3. Choose Codex.
  4. Describe the task you want Codex Agent to work on.

​
Choose a model

Choose GPT-6.1 Sol, GPT-6 Sol, or GPT-6 Luna from the Codex model menu for conversations and automations. Available models depend on your model provider, organization policy, and the agent version installed in the environment. Existing conversations and automations retain their saved model settings. These models offer Low, Medium, High, and Extra high reasoning. The defaults remain GPT-5.6 Sol, Medium reasoning, and Standard speed. Fast is available when your provider and organization policy allow it; for GPT-6.1 Sol, Fast costs 2x Standard. Bedrock supports Standard speed only.

​
Organization controls

Administrators can control Codex Agent behavior for the organization.

​
Model availability

In Organization Settings > Agents, administrators can allow or disable GPT-6.1 Sol, GPT-6 Sol, and GPT-6 Luna independently, limit reasoning effort, and control Fast mode.

​
Goal mode

Goal mode lets members set Codex goals that keep running until achieved. It is enabled by default. To change it, go to Organization Settings > Agents and toggle Goal mode in the Codex settings. When disabled, members can still run Codex Agent, but requests that ask it to run in goal mode are rejected.

​
Usage and billing

Compute and model usage follow your organization’s plan and model-access configuration:
  • On Enterprise, model usage follows the BYOK configuration for your runner.
See Cost & Budgets and Model access for details.
  • Model access: configure model access for AWS and GCP runners
  • AGENTS.md: add project context for agents
  • MCP servers: extend agent capabilities with external tools

​
Troubleshooting

Hard refresh the dashboard to load the latest model list. Check that your organization policy and model provider allow the model. An existing environment also needs a compatible agent version; availability in a new environment does not update the agent installed in an older one.