Repository navigation
**Title:** Support multiple model profiles in settings.json for quick switching via /model #204
Description
Activity
Proposal
I'd like to work on this feature.
My goal is to add support for named model profiles while keeping the current configuration fully backward compatible.
Design
- Introduce an optional
profilesfield insettings.json. - Each profile can override any model-related configuration, including:
modelbaseUrlapiKeythinkingEnabledreasoningEfforttemperature
- Add an optional
defaultProfilefield to select the active profile when the CLI starts. - If
profilesis not configured, the CLI will behave exactly as it does today.
Configuration Merging
To reduce duplication, profiles should act as overrides rather than requiring every field to be repeated.
Example:
{ "temperature": 0, "thinkingEnabled": true, "profiles": { "flash": { "model": "deepseek-v4-flash", "thinkingEnabled": false, "temperature": 0.7 } }, "defaultProfile": "flash" }The runtime configuration would be resolved by merging the selected profile over the global settings, allowing profiles to specify only the fields they need to override.
/modelUX- Display configured profiles in the
/modelmenu. - Selecting a profile immediately applies its configuration.
- If no profiles are configured, preserve the existing
/modelbehavior.
Backward Compatibility
- Existing
settings.jsonfiles continue to work without modification. - All new fields are optional.
- Users who don't use profiles won't notice any behavioral changes.
Implementation Plan
- Extend the configuration schema to support
profilesanddefaultProfile. - Update the configuration loader to resolve the active profile by merging it with the global configuration.
- Modify the
/modelcommand to display and switch between configured profiles. - Add validation and error handling for invalid or missing profiles.
- Add tests and update the documentation.
If this approach aligns with the project's direction, I'd be happy to implement it, including tests and documentation.
- Introduce an optional
Problem
Currently,
settings.jsononly accepts a single set of model-related configuration:{ "model": "deepseek-v4-pro", "thinkingEnabled": true, "reasoningEffort": "max", "temperature": 0 }When I need to switch between different models (e.g.,
deepseek-v4-prowith max reasoning for complex tasks,deepseek-v4-flashwithout thinking for quick edits), I have to manually adjust each field via/modelevery time — model name, thinking mode, reasoning effort, and temperature. This is repetitive and error-prone.Proposed Solution
Support a
profilesfield insettings.jsonto predefine multiple named configurations:{ "profiles": { "pro-max": { "model": "deepseek-v4-pro", "thinkingEnabled": true, "reasoningEffort": "max", "temperature": 0 }, "flash-quick": { "model": "deepseek-v4-flash", "thinkingEnabled": false, "temperature": 0.7 }, "coding-plan": { "model": "ark-code-latest", "baseUrl": "https://ark.cn-beijing.volces.com/api/coding/v3", "thinkingEnabled": true } }, "defaultProfile": "pro-max" }The
/modelcommand would then list these profiles for one-click switching.Use Cases
pro-maxflash-quickfor speed and cost savingscoding-planwithout re-enteringBASE_URLandAPI_KEYeach timeExpected Behavior
profilesentries accept:model,baseUrl,apiKey,thinkingEnabled,reasoningEffort,temperaturedefaultProfile(optional) specifies which profile to use at session start/modellists all profiles and allows selecting one to apply immediatelyprofilesis not configured,/modelbehaves exactly as it does todayThanks for the great tool!