The inlineChat.enableV2 controls how the chat extension computes edits for inline chat. This is tool based and uses less pre-processing/filtering before making request. This means responses are generally of higher quality but it also means that responses are often slower (due to the tool calling overhead of extra tokens and so). This item is about optimising the new backend without loosing its new powers

Some ideas

• disable intent detection
• use cache-breakoints
• show progress about editing
• in select cases use the good parts of the old approach
• use the prediction field
• don't always enable MultiReplaceString tool
• explore a replace_selection tool #280907
• better UI that makes waiting more pleasant