Area
Dashboard
What are you trying to accomplish?
I want to control whether synthetic --fast model rows appear in external client model pickers from the Dashboard. For clients such as Pi and DSH, the extra Model Fast entries add noise when I prefer to select the base model and control reasoning effort separately.
What prevents this today?
OpenCodex already supports the persisted fastRows configuration flag and preserves an explicit fastRows: false, but version 2.49.0 has no Dashboard control for it. The current workaround requires editing config.json manually and restarting the service, while Fast selectors are enabled by default.
What should OpenCodex do?
Expose a Dashboard switch for showing or hiding synthetic Fast selector rows. The control should reflect the effective fastRows value, persist changes through the existing configuration mechanism, and clearly state that it affects generated --fast picker rows only. Disabling it must not hide base models or change their supported reasoning-effort levels.
Example usage or interface
For example, under the model/catalog settings:
Show Fast model rows [On]
Adds eligible “Model Fast” selectors to external client model pickers.
This does not change reasoning-effort options on the base model.
Turning the switch off should produce the same effective configuration as:
After client integration refresh, entries such as gpt-5.6-sol Fast should disappear while gpt-5.6-sol remains available.
Alternatives or workarounds
Manually add "fastRows": false to config.json, restart OpenCodex, and refresh the affected client integrations. This works, but it bypasses the Dashboard workflow and is less discoverable and more error-prone than a switch for the existing setting.
Additional context
This proposal requests a UI surface for the existing behavior; it does not request a new Fast routing mode.
Checks
Area
Dashboard
What are you trying to accomplish?
I want to control whether synthetic
--fastmodel rows appear in external client model pickers from the Dashboard. For clients such as Pi and DSH, the extraModel Fastentries add noise when I prefer to select the base model and control reasoning effort separately.What prevents this today?
OpenCodex already supports the persisted
fastRowsconfiguration flag and preserves an explicitfastRows: false, but version 2.49.0 has no Dashboard control for it. The current workaround requires editingconfig.jsonmanually and restarting the service, while Fast selectors are enabled by default.What should OpenCodex do?
Expose a Dashboard switch for showing or hiding synthetic Fast selector rows. The control should reflect the effective
fastRowsvalue, persist changes through the existing configuration mechanism, and clearly state that it affects generated--fastpicker rows only. Disabling it must not hide base models or change their supported reasoning-effort levels.Example usage or interface
For example, under the model/catalog settings:
Turning the switch off should produce the same effective configuration as:
{ "fastRows": false }After client integration refresh, entries such as
gpt-5.6-sol Fastshould disappear whilegpt-5.6-solremains available.Alternatives or workarounds
Manually add
"fastRows": falsetoconfig.json, restart OpenCodex, and refresh the affected client integrations. This works, but it bypasses the Dashboard workflow and is less discoverable and more error-prone than a switch for the existing setting.Additional context
This proposal requests a UI surface for the existing behavior; it does not request a new Fast routing mode.
Checks