[Feature]: Manage Claude Desktop model visibility with persistent per-client selections #7702

Area

Dashboard

What are you trying to accomplish?

Choose which models appear in Claude Desktop without changing the models available to other OpenCodex clients. For example, keep Claude, Gemini and DeepSeek while hiding GPT, Grok and other families, using the dashboard rather than editing JSON.

What prevents this today?

Observed in OpenCodex v2.91.0 on Windows, at the Claude Desktop integration page:

• The Desktop model list has no per-model visibility checkbox or bulk include/exclude action. It includes unavailable entries and displays the full dashboard roster rather than clearly identifying the list currently applied to Desktop.
• Expanding Advanced, then a family, then a model exposes a default selector and a move-to-family action. These do not hide the model. Non-Claude models are grouped under Claude tiers, which makes model selection difficult to understand.
• The same status bar can simultaneously say that configuration changed and needs reapplication, and that configuration is up to date. These apparently refer to different states, but the UI does not explain the distinction.
• The page does not clearly explain whether applying its generated configuration will replace a manually curated Desktop list.

What should OpenCodex do?

Provide a persistent, client-specific model selection for Claude Desktop:

• Search and filter by actual model family/provider, with per-model visibility checkboxes and bulk selection.
• Keep model exposure separate from Claude tier mapping and default/quick-task roles.
• Show selected, available and currently applied counts separately; identify unavailable entries without including them ambiguously in the applied count.
• Preserve the selection across catalog refresh and configuration reapplication. Explain how newly discovered models are handled.
• Show a preview of added/removed models before applying, with clear saved-versus-applied status.
• Leave provider-wide selections and other clients unchanged.

Example usage or interface

Open Connections > Claude Desktop > Models. Filter for Claude, Gemini and DeepSeek, select the desired rows, and click Save and apply. Desktop then lists only those choices. Reopening the dashboard or refreshing provider catalogs preserves the choices. Advanced tier mapping remains optional.

Alternatives or workarounds

Editing the native Claude Desktop third-party configuration's inferenceModels list with modelDiscoveryEnabled set to false works around the long picker, but bypasses OpenCodex's generated profile and is difficult to maintain. Changing provider-wide model selections has a larger impact than this task requires. JSON import/export and moving models between Opus/Fable/Sonnet/Haiku are not substitutes for a visibility control.

Additional context

This is a model-list management request, not a request to change inference routing or account credentials. Related but different: #1213 concerns native/OpenCodex switching. The primary acceptance criterion is completing the selection in the dashboard and retaining it after reapplication, without hand-editing JSON.

Checks

• I searched existing issues and documentation.
• This request describes a concrete OpenCodex workflow rather than merely naming a desired technology.
• I removed secrets and personal data.