Role assignment and Tokens in GPT OSS vs Premium model

Summary:

Hi all,

I need help with two things:

1. Role assignment: When we create a new custom role with all the required AI privileges, the AI agent works as expected. However, when we assign the same AI privileges/permissions to an existing custom role that’s already in the system, the AI doesn’t work. Is this expected behavior, or are we missing any additional configuration?
2. Token limit: I need some clarification on what the token limit means. When selecting an LLM (for example, GPT OSS), it mentions a maximum of 4,000 tokens, while the premium models mention 32,000 tokens. Does this limit apply per request/response (conversation context), or is it an overall usage limit? For example, once 4,000 tokens are consumed, will the free model stop working, or does the limit reset for each interaction?