Connect your first AI client
Choose a supported setup mode, configure a client, and verify a real request.
Choose a supported setup mode, configure a client, and verify a real request.
Choose a client in the compatibility matrix. You need API access in the intended organization, available credits for the final request, and an unlocked operating-system credential store. Install a release of the UnlimGPT CLI that lists configure in its help, then check:
unlimgpt --version
unlimgpt --help
unlimgpt clients
The client list identifies automatic, secure-launch and guided setup. The steps below use Codex as the automatic example. For a different client, open its specific guide before changing its settings.
Preview the owned configuration changes without signing in, creating a key or writing files:
unlimgpt configure codex --dry-run
Read the destination and proposed changes. Quit Codex Desktop completely before applying them; closing only its window may leave the process running. If you have earlier Codex chats, the CLI backs up session stores before switching the active provider. That backup contains chat content and needs the same protection as your local chats. Then run:
unlimgpt configure codex
The CLI asks you to confirm its changes. If needed, it opens browser sign-in, creates a key for the selected profile and stores the secret in the operating-system credential store. A successful automatic setup prints a configured confirmation and an Undo: command for each file change. Save those receipt commands for a guarded revert; they do not revoke the API key.
unlimgpt whoami --json
unlimgpt models
The JSON response identifies the CLI OAuth account and organization. Compare data.organization.id with the organization selected in the browser. The model list checks the stored API key, without making a paid generation. These commands use different credentials: a successful model list alone does not prove the key belongs to the organization shown by whoami.
If setup printed a new API key ID, confirm that ID appears in unlimgpt keys list for the selected organization. If this CLI profile already held a key and you have switched organizations, stop before making a paid request. Run unlimgpt keys create --name codex-current-org while signed in to the intended organization; this selects the new key for the profile. Quit Codex Desktop and rerun unlimgpt configure codex, then repeat the checks. Review the old key separately in its original organization. If the OAuth organization itself is wrong, correct the CLI authorization.
Choose a canonical model ID from unlimgpt models, not its display name. If the configured model is absent, quit Codex Desktop and rerun configure with --model set to an ID from the list. For example, use unlimgpt configure codex --model gpt-5.6-sol only if the catalog lists gpt-5.6-sol. Save any new Undo: receipt before reopening Codex.
Reopen Codex Desktop, check that earlier chats are still visible, then start a new chat and ask for a short response with an available model. Wait for a non-empty answer; its wording may vary and the call may consume credits. Find the operation by endpoint, time and outcome in API Activity. If the client fails to connect, follow the Codex-specific guide and then run doctor for diagnosis. Doctor is not the success check.
Claude Code also supports automatic setup. Continue CLI uses secure launch; Cursor, Cline and Roo require guided steps in their native settings. In guided mode, configure explicitly says the client was not configured and gives the next steps; a clipboard key is a secret. Follow the client-specific guide and verify a real request there. Before repeating a failed setup, inspect its output and API keys: a key or receipt may already exist.