Skip to main content

Models and Providers

Choosing a Model​

The model name sits at the left of the panel header. It appears when you move the pointer over the panel. Click it to pick a different model.

Model dropdown in the chat header

  • Default: Lets Claude Code pick, using the model setting of your Claude account. Once you have sent a message on Default, the row shows which model that is, for example Currently Fable.
  • The models you can pick, with a short description of each. The list is refreshed from Claude Code after your first reply.

You can switch models in the middle of a chat. The new model takes over from your next message and keeps the conversation history, and a notice in the chat confirms the switch. The first reply after a switch can take a little longer. The message box placeholder shows which model you are talking to, like Ask Opus 5.5....

The model dropdown belongs to Visual AI. In a CLI session, pick the model inside the CLI.

Thinking Effort​

Some models can think more or less before they answer. For these, a bulb chip sits next to the permission mode at the bottom of the panel. Click it to pick a level, from Low for quick, simple tasks to Max for hard problems that need deep reasoning. Higher levels take longer and use more tokens. Default leaves the choice to the model.

Thinking effort

The levels offered depend on the selected model, and your choice is remembered per model. The chip is not shown for models without this setting, or when a provider with a custom endpoint is active.

Settings​

Click the gear button in the panel header, or the AI Settings link on the start screen, to open the settings dialog.

AI Settings dialog

The dialog has a tab for each CLI: Claude Code, which also covers Visual AI, and Codex. Each tab has:

  • Active Provider: Which provider to use. Default (Local Claude Code) uses Claude Code's own login, and Default (Local Codex) does the same for Codex. A provider on the Claude Code tab applies to Visual AI and to Claude Code CLI sessions started from the panel.
  • Providers: The custom providers you have added, with Edit and Delete buttons.
  • Path to the executable: Where the CLI is installed. Leave it blank to let Phoenix Code find it.
  • Connect to Phoenix: Whether a CLI started from the panel is connected to the editor. See Claude Code CLI and Codex CLI.

At the bottom, Enable AI features switches AI on or off. See Turning AI Off. Click Done to save your changes, or Cancel to drop them.

Adding a Provider​

Click + Add Provider to use your own API key or endpoint:

Add provider form

  • Name: Any name you like.
  • API Key: Your key. It is stored encrypted on this computer.
  • Base URL: The endpoint of the provider. Leave it blank to use the provider's default.
  • API Timeout (ms): How long to wait for a reply. Claude Code only.

Click Save Provider, then pick it under Active Provider and click Done. When a provider with a base URL is active, the chat shows a one-time Using custom API endpoint notice on your next message.

With a provider active on the Claude Code tab, Visual AI skips the Claude login.

Compatible Providers​

  • Anthropic Claude (default): Recommended. Create an API key at platform.claude.com.
  • z.ai GLM: Tested working as a drop-in alternative. Use z.ai's Anthropic-compatible endpoint as the base URL.
  • Other Anthropic-compatible providers: Add the provider's base URL and API key on the Claude Code tab.
  • OpenAI-compatible providers: Add them on the Codex tab. They apply to the Codex CLI.