Compare

The One vs Codex

Choose Codex for coding with OpenAI's models in ChatGPT, your terminal or your editor: local runs are sandboxed by default, cloud tasks keep going while your computer sleeps, and it reviews pull requests on GitHub. Choose The One to hand ongoing work to teams of agents that can mix vendors, with each matter on a board and deliveries you judge. The two also work together: Codex can call The One's working groups through The One's MCP server, and The One can use Codex models through your ChatGPT sign-in.

What it isA desktop app: a main assistant hands bigger work to working groups of AI agents, each matter on its own boardOpenAI's coding agent, the developer mode of ChatGPT
Where you use itThe desktop app; a Linux server for headless runsThe ChatGPT desktop app (Codex mode), CLI, VS Code, Cursor and Windsurf, the web and mobile apps, Slack, Linear
PlatformsmacOS 15+ on Apple Silicon, Windows, and a Linux servermacOS, Windows and Linux (preview) desktop app; CLI on macOS, Linux and Windows
ModelsAny vendor, chosen per agent, including Codex models through a ChatGPT sign-inOpenAI's GPT-6 models; other providers only if they speak the Responses API; local models through Ollama or LM Studio
PriceNo charge with your own keys or ChatGPT account; membership $20 / $60 / $200 a month with model usage includedIncluded in ChatGPT Plus $20, Pro $100 / $200 / $500 and Business; limited on Free and Go; or an API key (CLI and IDE only)
Several agentsWorking groups from templates: roles, handoffs, parallel agents, a worker-and-reviewer task board, sub-groups, lasting across daysSubagents (on by default; custom agents share the parent's provider); parallel chats in worktrees; /goal for long work
UndoAn action log of every tool call; undo a call, a run, a matter or everything since a timeNo file checkpoints; the docs suggest Git checkpoints, and the app reverts a diff, file or hunk through Git
When your computer is offAgents stop, unless they run on your Linux serverCloud tasks run in their own VMs
AutomationTriggers on a schedule or cron, when a folder, a web page or a mailbox has something new, or on a webhookApp schedules while the computer is on; cloud schedules with Gmail, Slack and GitHub triggers; GitHub Action and @codex review
From your phoneAgent Room channel, textThe ChatGPT mobile app starts, steers and approves tasks
SourceClosed source; the docs are publicCLI and SDK Apache-2.0 (no outside contributions); app, IDE extension and cloud closed

Choose Codex if

  • You code mostly with OpenAI's models and already work in ChatGPT, a terminal or an editor.
  • You want commands sandboxed and approvals handled before they run.
  • You want cloud tasks that keep going when your computer is off, and pull-request review on GitHub.

Choose The One if

  • You hand over work that runs for days and want it on a board, coming back as deliveries with evidence.
  • You want a team that mixes models from several vendors, not one provider per session.
  • You want to hand work off and come back to a delivery, rather than steer each task yourself.

Using them together

  • Codex can call The One's working groups through The One's MCP server on your computer.
  • The One can use Codex models with your ChatGPT sign-in, drawing on your Plus or Pro plan's Codex allowance.

Questions

Does The One replace Codex?
No. Codex is for hands-on coding with OpenAI models; The One is for handing work to teams, and it can take jobs from Codex.
Can The One use OpenAI models?
Yes, with an OpenAI API key, or with your ChatGPT sign-in for Codex models.

Facts checked on 2026-11-13 from learn.chatgpt.com and developers.openai.com, and from The One's docs. Something changed? Tell us at support@the-zeroth.com