Use Claude or ChatGPT through the CLI you already use
chekd can run optional model-backed actions through Claude Code or Codex after it verifies that CLI’s account sign-in and you choose that exact provider. chekd adds no per-use model fee. Your provider plan, usage limits, charges, and terms apply.
How this works
Claude and ChatGPT appear as separate cards. Finding one of their CLIs is not consent, and chekd never picks the only detected provider for you. A card becomes usable only after the CLI reports a supported account session; choosing Use Claude or Use ChatGPT is the consent action.
That exact provider is remembered locally. chekd rechecks its sign-in before use. If the sign-in is no longer ready, the model-backed action pauses instead of silently switching to another cloud provider; model-free walking and review continue.
Your CLI. Your account. Your call.
Set it up
- Install and sign in through the provider Use the provider’s current official setup: Claude Code setup or Codex CLI setup. chekd never asks for or stores that account credential.
- Return to chekd Open the model picker again. A ready account gets its own provider-specific card; a CLI that is installed but not ready still points to setup.
- Choose the exact provider Select Use Claude or Use ChatGPT. Nothing cloud-backed is selected merely because a CLI, marker file, login, or provider key exists.
Prefer not to choose a model? Skip setup. The objective browser walk and finding review still work without one.
What leaves this machine
chekd stores findings, screenshots, corrections, and run history locally. When a model-backed action runs, it sends the context that action needs through your selected CLI to the provider. Depending on the action, that can include page text, finding details, source snippets, and screenshots. The provider handles that data under its policy; chekd does not upload it to a chekd service.
“Rides your subscription” is not a $0 guarantee. Provider quotas, credits, overages, and billing rules can change; check your provider plan. Local Ollama uses your own device resources and keeps app context on this machine, with no provider per-use bill. Antigravity is a separate text-only Experimental option that must be enabled explicitly in AI & governance; source snippets need another grant and screenshots stay blocked. Paid API keys are a working last resort with explicit authorization, a capped connection test, and per-job spend ceilings.