Model access
Where a Cmdop conversation’s inference comes from follows the execution engine you chose. The built-in CMDOP engine uses the CMDOP Router and needs a platform API key on the host before the web console unlocks. Claude Code and Codex use their own existing local sign-ins, which Cmdop drives but never collects, stores, or manages.
Who pays for and provides inference follows the execution engine selected for a conversation.
Engine ownership
| Engine | Inference | Credential |
|---|---|---|
| CMDOP | The CMDOP Router | Your platform API key, held by the host |
| Claude Code | The local Claude Code CLI | That CLI’s existing account |
| Codex | The local Codex CLI | That CLI’s existing account |
The CMDOP engine needs a key before you start
The first time you open the web console, the relay authenticates your browser and then checks for an inference key. If none is present, a Connect your CMDOP account step blocks the product shell until you paste a platform API key or complete the account handoff.
The key is stored in the host’s secret store and is never echoed back by status APIs. You can replace it later under Settings → AI Providers.
This key is separate from every other credential in Cmdop — it gates inference, not machine access. See Identity and access.
External engines bring their own
Claude Code and Codex use their own accounts on the machine where they run. Cmdop drives the installed CLI and never collects or transfers those credentials. Choosing a machine or a project does not hand an external engine’s account to Cmdop, and configuring a CMDOP key does not pay for their usage.
Engine and model are separate choices
Selecting a model changes how an already-chosen engine reasons. It does not change the machine, the project, or the session’s identity. With CMDOP, the model is chosen per turn; with the driven CLIs, the model binds when the session begins.
Your plan sets which models you can pick
With the CMDOP engine, your plan determines the models available to you. The model picker shows more capable models as locked, with the plan that unlocks them; higher plans reach stronger models. Choosing a locked model does not silently fall back — the turn reports that the model is not included in your plan and points you to pick another or upgrade. This is a capability ceiling, not a per-message charge: Cmdop bills by machines, not by tokens (see Pricing).
Related
Common questions
Which credential pays for inference?
It depends on the engine. The built-in CMDOP engine uses your platform API key; Claude Code and Codex use their own local CLI accounts.
Does a CMDOP API key pay for Claude Code or Codex?
No. Configuring a CMDOP key does not pay for external CLI usage, and choosing Claude Code or Codex does not hand their account credentials to Cmdop.
Can I change the model without changing the machine?
Yes. Model choice changes how the selected engine reasons; it does not change the machine, project scope, or session identity.
Can I use any model on any plan?
With the CMDOP engine, no — your plan sets a model ceiling. More capable models appear locked in the picker until your plan includes them, and a request for a locked one is refused rather than downgraded. The driven CLIs use their own accounts, so their model choice follows those accounts, not your Cmdop plan.
Cmdop