Describe the feature or problem you'd like to solve
Currently, the Copilot CLI uses the primary chat model to handle auxiliary tasks like context compaction. I propose decoupled model configuration. This would allow users to map specific sub tasks, such as summarization/compaction and task agent execution to different models based on the difficulty of the required task. For example: I would like to use a lightweight model for compaction but a more knowledgable model for task execution/exploring.
Proposed solution
Introduce a configuration section (in config.json) that allows mapping tasks/default subagents types to specific models.
Example prompts or workflows
No response
Additional context
No response
Describe the feature or problem you'd like to solve
Currently, the Copilot CLI uses the primary chat model to handle auxiliary tasks like context compaction. I propose decoupled model configuration. This would allow users to map specific sub tasks, such as summarization/compaction and task agent execution to different models based on the difficulty of the required task. For example: I would like to use a lightweight model for compaction but a more knowledgable model for task execution/exploring.
Proposed solution
Introduce a configuration section (in config.json) that allows mapping tasks/default subagents types to specific models.
Example prompts or workflows
No response
Additional context
No response