Selecting the session mode (and model) before it starts

[arthurcgusmao]

I have been experimenting with opencode and agent-shell.
What I'm missing is the option to modify the model and the mode, especially the latter, before sending any request to the LLM.

For example, say I want to plan the implementation of a new feature. I start agent-shell and press C-c RET, only to get:

agent-shell-set-session-mode: No active session

Am I missing something, or is this yet to be implemented?

Thanks a lot for the great package!

Checklist

• I've read the README's Issues section.
• I'm running the latest agent-shell version.
• I'm running the latest agent version (ie. Claude Code, Geminig CLI).
• I'm running the latest agent ACP package (ie. claude-code-acp).
• For requesting new agent support, I'm including a link to the ACP-capable agent or related ACP package.
• For issues, I'm including ACP traffic (as per README).

[vytas7]

Same here, would be nice to choose before the start and/or configure a default model only for agent-shell.
So far, I have partially worked this around by setting an Opencode-wide default model in its config, e.g.:

{
    "$schema": "https://opencode.ai/config.json",
    "model": "ollama/<something>",
    ...
}

[xenodium]

agent-shell-set-session-mode: No active session

We currently have a limitation. We aren't able to set these until after sending the first prompt, which triggers a handshake to find out what is possible by the agent. Having said that, I've just introduced two default variables in 812d17e you can use for OpenCode default values.

Can you give that a try? You'll need to update to the latest: v0.34.1.

[vytas7]

Can you give that a try? You'll need to update to the latest: v0.34.1.

Thanks, 20260208.1246 from MELPA works fine for me (wrt model selection)!

[xenodium]

Thanks, 20260208.1246 from MELPA works fine for me (wrt model selection)!

Glad to hear. If you're getting value out of agent-shell, please consider sponsoring to make this work sustainable. Keep in mind I'm not an OpenCode user, so any support I provide or changes are purely to help with your requests.

[vytas7]

@xenodium we'll see where my Emacs setup lands in a longer term wrt the "AI" age -- if I keep using agent-shell, I'll definitely consider supporting financially too.

[xenodium]

Makes sense. If looking to learn more about agent-shell flows/features, I demoed most of the main features in this video https://xenodium.com/bending-emacs-episode-10-agent-shell

[arthurcgusmao]

I've just introduced two default variables [...]

Thanks a lot @xenodium !

We currently have a limitation. We aren't able to set these until after sending the first prompt, which triggers a handshake to find out what is possible by the agent.

Got it, is it a limitation on ACP itself, or something on Emacs end?

[xenodium]

It's a limitation on agent-shell for the time being. When you first start a shell, you haven't yet communicated with the agent, so you don't know its capabilities (including whether or not you can set mode or model). I have been considering moving away from comint-mode (parent mode), to have more flexibility.