Another provider

Each pane can run on a different provider (native Claude, GLM/z.ai, or any other Anthropic-compatible host) and the IDE tools work the same either way. What makes that possible is a profile.

What a profile is

A named set of environment variables (e.g. ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN, model overrides) injected into that pane’s claude.exe, so a pane can run on a different provider while the IDE MCP tools keep working.

Profiles are edited under Tools → Options → cv4vs Agents → Profiles. It is not a settings table but an editor:

Profiles are listed on the left (the checkbox enables one), and edited on the right: a name, optional notes, and the environment grid, pre-filled with the keys you are most likely to need, so a new profile is usually just a matter of pasting two values. Available environment variables links to Anthropic’s reference for everything else the CLI understands.

Enabled profiles appear under View → cv4vs Agents, and the active one is shown in the pane caption and toolbar.

A few rules:

• Claude, the native profile, is not stored: it is the process environment as it is. A profile of your own named Claude replaces it.
• Setting CLAUDE_CONFIG_DIR in a profile gives it its own config directory: separate login, settings, sessions, statistics and file backups. Profiles without it share ~/.claude.
• A new pane opened from a pane’s toolbar inherits that pane’s profile.
• A profile with a blank name, or disabled, is ignored everywhere.
• A profile’s variables are applied first and the extension’s own last, so a profile cannot override CLAUDE_CODE_ENTRYPOINT.

Paste from JSON

The editor’s Paste from JSON button (the clipboard icon under the grid) fills the env grid from the clipboard, so you can lift a provider’s snippet straight from its docs. It accepts either a full settings block:

{ "env": { "ANTHROPIC_BASE_URL": "…", "ANTHROPIC_AUTH_TOKEN": "…" } }

or a plain key/value map:

{ "ANTHROPIC_BASE_URL": "…", "ANTHROPIC_AUTH_TOKEN": "…" }

The env object is used when present, otherwise the whole object.

Provider setup guides

• z.ai / GLM: GLM, Kimi, DeepSeek, Qwen, MiniMax
• Qwen: DashScope Anthropic API
• MiniMax: Anthropic-compatible models
• DeepSeek: direct Anthropic API compatibility
• OpenRouter: multi-provider gateway
• Ollama: local open-source models
• Complete alternative models guide: comprehensive provider list

Without a profile

Profiles are not the only way: the CLI reads these variables from the process environment like any shell would, so setting them at the OS level works too. Profiles are usually preferable: one pane per provider, switchable without touching your system environment.

What changes on another provider

The IDE tools can go missing

Pointing ANTHROPIC_BASE_URL at a custom host can disable the IDE MCP tools. That is a CLI-side restriction, not something the extension controls.

• Plan usage. A profile that sets an API key, an auth token, a base URL, or the Bedrock or Vertex switch has no plan limits, so the status bar item shows the icon alone, dimmed.
• Remote Control isn’t available on Amazon Bedrock, Google Vertex AI, or Microsoft Foundry, and not with a custom ANTHROPIC_BASE_URL: see Remote Control.
• Statistics show model names exactly as the API returned them, so a third-party provider’s model ids stay readable.

Where the token is stored

Caution

profiles.json holds the environment variables you enter in the Profiles page, ANTHROPIC_AUTH_TOKEN among them. It is a plain file with no encryption, readable by anything running as your user.

Unlike General, Chat and Debug, profiles are not stored in the VS settings store: they live in profiles.json so the menu can list them without opening the Options page first; see Settings and data.