I'm testing the new AI Assistant on FM 27 using qwen3.5-9b on LM Studio as the provider but it's constantly giving the error "The assistant reached the limit of eight tool rounds. Send another message to continue." and then locking out the conversation.
As far as I know it's a problem with the AI model itself and not FMRTE, but maybe it can be fixed? Not sure.

Anyway, mostly just making this post to ask if anyone has any local models they recommend for the assistant that actually work well (and fast)?

+Context, here's the error output in LM Studio when the tool round limit is reached in a chat

[2026-10-31 07:19:04][ERROR][qwen/qwen3.5-9b] Engine protocol predict request returned 500: {"error":{"code":500,"message":"\n------------\nWhile executing CallExpression at line 85, column 32 in source:\n...first %}↵ {{- raise_exception('System message must be at the beginnin...\n ^\nError: Jinja Exception: System message must be at the beginning.","type":"server_error"}}. Error Data: {

"status": 500,

"responseBody": {

"error": {

"code": 500,

"message": "\n------------\nWhile executing CallExpression at line 85, column 32 in source:\n...first %}↵ {{- raise_exception('System message must be at the beginnin...\n ^\nError: Jinja Exception: System message must be at the beginning.",

"type": "server_error"

}

}

}, Additional Data: n/a

Can you please give all the info?

@BraCa Some AI problems

The FM27 version is 27.0 Game Build 26227640 (Default Public Version) on Windows 11.
FMRTE Version 27.0.0-build3, running as administrator with no license.

The issue seems to be when you ask the model to use any player's statistics, it seems to not retrieve any rows for "get_player_stats"? A lot of activity lists seem to have this orange warning icon



Also here's my model details if you need them:

I've been experimenting a bit with Ollama today. I tried using llama3.2:latest and this model actually gives me results @Rubix. Not sure if this is the best model to use though, I'm completely new to local AI models