Am I missing a way to use multiple models?

[kfsone]

Perhaps because of spending time tinkering with MS Autogen a year+ ago, I have this intuition to use different models for different parts of the boomerang process. For instance, Claude does great at Orchestrator and Code, but the debug role/prompts will empty your token balance in no time(*)

In particular, I'd like to be able to set up coding pairs - I have a couple of local machines with a gemma3/deepseek combo that riff very well together at writing code and reviewing it before accepting it, and the latest granite for testing the results.

Claude -> orchestrator master task,
| Claude -> sub-task driver,
| | Deepseek -> Coding/Producer, Sub Task 1
| | | Gemma3 -> Code author
| | Deepseek -> Coding/Producer,
| | | Granite -> Software test
| Claude ...
| | Deep ... Sub Task 2

(* Claude Sonnet is hell bent on writing code. It doesn't handle the do-not instruction well; it generates the code anyway, "I can't write code to a file, but here is the code I would if I could" [paraphrase]. Then, it'll document the code to a md/text file, then write its own summary including a repetition of the code. Orchestrator will take that and pass it to a code mode, giving the code mode a task AND telling it the code. Then, code mode will want to review everything, and will start by reading the documentation that contains another copy of the code, and then, code mode conclude the code has already been written and task-complete)

[samhvw8]

Hi @kfsone,

Each mode remembers your last-used model.
You can read more here: https://docs.roocode.com/basic-usage/using-modes?_highlight=remembers

You can configure your API Configuration profile in this section (Prompt setting):

[kfsone]

Ahh - thank you -- when I'd tried this a week or so ago, everything kept using the current settings, and debug in particular is a problem with claude.