Let a cheaper LLM fix tool call issues instead of retrying with the same provider

[CaliLuke]

Gemini makes a lot of tool call mistakes, and the constant round tripping can get expensive. Instead it would be cheaper and faster if a local llm or a cheaper model like 2.0-flash (or a different llm like Claude) could take care of those errors until the edit succeeds, then switch back to the original llm to continue the job.

[nickchomey]

Great idea

[Xytronix]

Please run a local llm that does the job