Describe the feature or problem you'd like to solve
Some of the models I have access to support multiple context window sizes. I want to start with the smaller context window, then upgrade to the next larger context window size for the same model type instead of compacting when the context fills up.
Proposed solution
Add a configuration setting with similar placement to "continueOnAutoMode" - instead of triggering on rate limiting and switching to auto model, trigger on full context (which would normally compact) and switch to larger context version of same model if available, only compacting if already on the largest context available for that model.
Example prompts or workflows
This would allow starting with a smaller context and seamlessly upgrading within the session if needed. Currently if I start with the smaller context and hit the limit, I have to manually switch then additionally prompt to attempt recovery of the context that was lost by compaction.
Additional context
Under the premium requests model, smaller context was typically half the multiplier of larger context - I'm assuming that AI credits work similarly. If not, and larger context model scales credit usage based on the actual context used instead of the max, then this feature is unnecessary as I can just start with the larger context every time.
Describe the feature or problem you'd like to solve
Some of the models I have access to support multiple context window sizes. I want to start with the smaller context window, then upgrade to the next larger context window size for the same model type instead of compacting when the context fills up.
Proposed solution
Add a configuration setting with similar placement to "continueOnAutoMode" - instead of triggering on rate limiting and switching to auto model, trigger on full context (which would normally compact) and switch to larger context version of same model if available, only compacting if already on the largest context available for that model.
Example prompts or workflows
This would allow starting with a smaller context and seamlessly upgrading within the session if needed. Currently if I start with the smaller context and hit the limit, I have to manually switch then additionally prompt to attempt recovery of the context that was lost by compaction.
Additional context
Under the premium requests model, smaller context was typically half the multiplier of larger context - I'm assuming that AI credits work similarly. If not, and larger context model scales credit usage based on the actual context used instead of the max, then this feature is unnecessary as I can just start with the larger context every time.