Totally lost with the 0.0.31, seems to be great feature wise, but how to configure agents/tools in settings?

[gsgxnet]

A few days ago I did the upgrade to zee 0.0.31 version of the llama-vscode extension. At first it looked like the new option to use agents within would be a great new feature. But my setup is so totally different from what seems to be the new defaults, that I have no clue how to get it going.
The wiki is no help for that because it is for the 0.0.27 version only.

I am lucky to have two modern Macs available, one M Mini which permanently serves Qwen3 Coder 30B using llama-server started by a LaunchDaemon within my private network. And one MacBook Pro which is setup to serve gpt-oss 120B locally also by a LaunchDaemon.
There is no need for me to configure any llama-server starting setups within the llama-vscode extension. As these services are used for other AI usage patterns here, there is no option to dynamically load and unload other models. And caused by memory constrains there would be the only option for the llama-vscode extension to start more models if they are tiny. Should work for the embeddings. But not for agents, I assume.

I understand how to configure these Chat and FIM services in the extension settings, looks like same as before. But I can not find out how to config the agent model. There are mostly references for modifying the settings.json file directly. But those 616 loc are too much for me to comprehend.

Could somebody write down how I could get the agents going with my, from my point of view, quite simple and powerful setup?

[igardev]

@gsgxnet OK, you have great machines :). Shortly - you just need to select tools model (similar as Fim and Chat) and the agent will work.

1. If you don't want to configure llama-server locally - when asked to install llama.cpp answer with cancel and on the next popup select Yes. After that you will not be asked to install/upgrade llama.cpp periodically
2. Agent is for now only system prompt + tools, not model. It uses the currently selected tools model.
3. You can add your tools models from llama-vscode menu (Ctrl+Shift+M) -> Tools models...->Add external tools model... (the name and the endpoint is needed in your case)
4. Once the model is configured - select it from llama-vscode menu (Ctrl+Shift+M) -> Tools models...->Select/start tools model... (or even easier from current environment view)

With this you should be able to use the agent.

In agent view you could change the tools model (the one used by the agent) and also you could choose some predefined agents (for now default, help and ask). With the help agent you could get some help about llama.vscode. With default - you could even change the source files, with ask - just ask about your code without changing the files.

If you need more details - don't hesitate to ask more questions.

[igardev]

@gsgxnet I guess in your current configuration you use Qwen3 Coder 30B for FIM and gpt-oss 120B for chat and you expect to set the endpoint for gpt-oss 120B also in setting endpoint_tools and the agent to work. This doesn't work now, but will be fixed in version 0.0.33 (I hope afer 1-2 days).