You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Hide custom models from discovery without changing saved selections
#29735
I use CLIProxy to expose a large model catalog to multiple harnesses. I want every model to remain configured there, but only see the models I normally use in T3 Code. Deleting a custom model just to remove it from a picker loses the configuration I want to keep.
T3 Code already has per-provider-instance model visibility controls. This proposal extends them to custom models, with a clearer contract: hide from discovery, do not disable execution or rewrite saved selections.
Proposed behavior
Where a custom model is referenced
What hiding it should do
Running turns and queued work
Nothing. Do not cancel work or substitute a model.
Existing threads and saved drafts
Retain the model and its options. Show its correct name, capabilities, and hidden status.
Environment/project defaults
Keep the explicit default, including for new threads.
Configured rate-limit fallback
Keep using it when the fallback applies. Resolve display metadata after the effective model is chosen.
Title/branch generation, source-control writing, and review guides
Retain explicit selections and existing inheritance.
Saved scheduled tasks
Preserve execution and display the saved model accurately when editing.
Ordinary picker choices, favorites, search, and MCP model discovery
Omit hidden models, even when one is currently selected.
Explicit known references from agents
Remain valid; hiding is not an authorization boundary.
Last-used model in a genuinely new thread
Skip it if hidden. Do not confuse this convenience hint with a saved draft or explicit default.
Automatic selection without a saved reference
Choose a visible candidate. If none exists, require a model to be shown or explicitly selected rather than silently picking a hidden one.
The key distinction is that selected-model metadata comes from the configured catalog; selectable choices come from the visible catalog. Keeping the label correct should not require putting a hidden selection back in the dropdown. After switching away, it stays unavailable as a new choice until shown again.
Scope
Reuse the existing visibility preference and bulk controls; no new settings schema or proxy/harness changes.
Keep custom definitions, capabilities, ordering, and favorites. Hide/show is reversible and isolated to each provider instance.
Honor the same preference on web/desktop, mobile, and MCP discovery.
Leave existing models visible by default.
This is about custom-model visibility, not a general redesign of saved hidden built-in selections. Existing built-in selection-resolution rules are not being replaced with a new universal retention policy.
Contribution
The initial implementation in #29743 was closed because it mixed selected metadata with selectable options and left these product decisions implicit. The current implementation is #29924, which replaces #29833 with the same behavior rebuilt as a commit-by-commit series on current main. Its description has the behavior matrix and screenshots. This discussion stays open for maintainer feedback; a linked implementation is not a claim of maintainer approval.
This is intended as a focused configuration extension of the existing model-visibility capability, rather than a new provider workflow. Would this scope and distinction between saved references and automatic suggestions be acceptable?
Related work: #10947 added bulk model visibility; #29072 discusses broader OpenCode catalog grouping. Neither proxy-side filtering nor grouping/presets is part of this proposal.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Problem
I use CLIProxy to expose a large model catalog to multiple harnesses. I want every model to remain configured there, but only see the models I normally use in T3 Code. Deleting a custom model just to remove it from a picker loses the configuration I want to keep.
T3 Code already has per-provider-instance model visibility controls. This proposal extends them to custom models, with a clearer contract: hide from discovery, do not disable execution or rewrite saved selections.
Proposed behavior
The key distinction is that selected-model metadata comes from the configured catalog; selectable choices come from the visible catalog. Keeping the label correct should not require putting a hidden selection back in the dropdown. After switching away, it stays unavailable as a new choice until shown again.
Scope
Contribution
The initial implementation in #29743 was closed because it mixed selected metadata with selectable options and left these product decisions implicit. The current implementation is #29924, which replaces #29833 with the same behavior rebuilt as a commit-by-commit series on current main. Its description has the behavior matrix and screenshots. This discussion stays open for maintainer feedback; a linked implementation is not a claim of maintainer approval.
This is intended as a focused configuration extension of the existing model-visibility capability, rather than a new provider workflow. Would this scope and distinction between saved references and automatic suggestions be acceptable?
Related work: #10947 added bulk model visibility; #29072 discusses broader OpenCode catalog grouping. Neither proxy-side filtering nor grouping/presets is part of this proposal.
All reactions