Provider, model, and voice problems

API key fails

Verify the provider, key scope, endpoint, and whether a custom provider expects different headers or compatibility settings. Saving a key replaces OAuth for that provider. Reauthenticate rather than placing a key in prompts or logs.

OAuth cannot complete

Only one callback-server flow can run at a time for selected providers. Cancel the active flow and restart. Corporate TLS/proxy failures can invalidate a PKCE-bound code; do not reuse it. Follow install/proxy guidance and begin a fresh flow.

Anthropic subscription authentication may use paid extra usage outside plan limits. This warning is expected, not an authentication error.

Expected model is missing

Authenticate the correct provider and clear/review model scope. Stale scoped entries show as unavailable and are ignored. Some provider catalogs expose only the first eight models. Define a compatible custom model if appropriate.

Thinking options are model metadata-dependent. A missing xhigh or max generally means the selected model does not declare it.

Voice is unavailable

Voice requires OpenAI Codex/ChatGPT OAuth with account information, not an OpenAI API key. Confirm microphone permission, keep recordings under 8 minutes/25 MB, and use a supported browser audio type.

Image request fails

For a text-only primary model, open Settings → Tools → Image explanation, choose a currently available image-capable model, then enable explain_image. If the configured model or its credentials were removed, the setting is marked unavailable and the tool is hidden until reconfigured.

The tool is intentionally absent when disabled or when the primary model already supports image input. It accepts JPEG, PNG, GIF, and WebP files up to 20 MB, detected from file contents rather than the extension.

GitHub PRs are unavailable

This is not provider auth. Install gh, ensure a GitHub remote exists, and run gh auth login.