awml / platforms / codex-agent-laptop
Codex on a laptop
Codex on a person's own laptop, started on a schedule by the ChatGPT desktop app's scheduled tasks on a local project, or by cron or launchd running `codex exec`. This describes a kind of platform, not anyone's machine; AWML's example codex-laptop is a different one, where the laptop runs the workflow's own code. Checked against OpenAI's documentation on 22 October 2026.
- Off for up to
- 14 days
A laptop may be shut for a fortnight's holiday, as in the AWML spec's own example. Scheduled tasks on a local project need the computer on and the desktop app running. A laptop that never sleeps can say less. https://github.com/chrismdp/awml/blob/main/SPEC.md#placing-a-workflow-where-it-can-run https://learn.chatgpt.com/docs/automations
- Runs steps as
- an AI agent
Each run is a Codex session that carries out the task's prompt. https://learn.chatgpt.com/docs/automations https://learn.chatgpt.com/docs/non-interactive-mode
- AI from
- openai
Codex's own agent, signed in with ChatGPT or an OpenAI API key. Routed through Amazon Bedrock, it is processed on Amazon's terms instead: list amazon here then. https://learn.chatgpt.com/docs/auth https://learn.chatgpt.com/docs/amazon-bedrock
- Can call
- none
Codex runs commands in a sandbox with network access off by default, and `codex exec` starts read-only. So the workflow's own code cannot call a model company until you turn on sandbox_workspace_write.network_access, or choose full access: say any here then. https://learn.chatgpt.com/docs/agent-approvals-security#network-access https://learn.chatgpt.com/docs/non-interactive-mode
- Holds
- notes
A laptop holds its owner's folders. A shared description cannot know their names, so it says notes, as AWML's examples do: keep a workflow's contexts under notes, or name your own folders when you describe your own laptop. https://github.com/chrismdp/awml/blob/main/SPEC.md#placing-a-workflow-where-it-can-run