Default Gemini CLI startup should pin a faster model

[Joseph19820124]

When OpenAB starts Gemini CLI without an explicit --model argument, Gemini CLI appears to default to a Pro model in practice.

That leads to noticeably longer thinking time and slower response latency for interactive chat usage.

Suggested change:

args = ["--acp", "--model", "gemini-3.1-flash-lite-preview"]

Using a faster Flash-Lite model by default would likely improve responsiveness a lot for Discord-style interactions, while still keeping ACP integration intact.

At minimum, it would be helpful to document the current default behavior and provide a recommended low-latency Gemini configuration in the README or example config.

[CHC-Agent]

Hi @Joseph19820124,

This is Gemini CLI's default model selection, not an OpenAB bug — users can already configure it via args in config.toml:

args = ["--acp", "--model", "gemini-3.1-flash-lite-preview"]

PR #177 is up to document a recommended config. One concern: hardcoding a specific model name in docs may go stale as Gemini releases new models.

Related: #185 / PR #186 adds a /model slash command for runtime model switching. If that lands, users could switch models on the fly without touching config — a more flexible long-term solution. The two approaches complement each other (docs for sensible defaults, /model for runtime flexibility).

[chaodu-agent]

This makes sense as a config improvement. Before we prioritize, could you share a bit more context?

1. Use case: What kind of Discord interactions are you running where latency matters most? (e.g. quick Q&A, code review, multi-turn debugging)
2. Observed latency: Roughly how long are responses taking with the default Gemini model vs. what you'd expect with Flash-Lite?
3. Have you tested gemini-3.1-flash-lite-preview with OpenAB? If so, any quality tradeoffs noticed?

This will help us decide whether to change the default or just document the recommended low-latency config.