Custom System Per-Model in Override Parameters

Right now, under Override Global Model Parameters, we can only set a single global system prompt that applies to all models. This is very limiting because:

• Each model often needs a different optimized system prompt to perform at its best.

  • Example: For GPT-5, I might set the system prompt to Knowledge cutoff: 2024-06.

  • For Claude, I would want Knowledge cutoff: 2025-01.

  • For GPT-4.1, I may use a more optimized or simplified structure that works better for that model.

• Using one global prompt for every model is not ideal, some prompts actually make performance worse on other models.

• While it’s possible to create separate agents with different system prompts, that’s not practical because it means creating and managing a new agent for every single model variation.

Proposed Solution

• Add a Custom System Prompt field per model in the Override Parameters section (where context limit, max tokens, reasoning effort are already set).

• Handle edge case when a user switches models mid-conversation, the system prompt should automatically change to the one defined for that model without needing to restart the conversation.