memory_commit is currently hardwired to use Antigravity for its subagent, regardless of my active chat model.
Specifically:
- src/subagent.ts
COMMITTER_MODEL = "google-antigravity/gemini-3-flash"
- agents/memory-committer.md
model: google-antigravity/gemini-3-flash
So even if my main session has valid creds (Anthropic/OpenAI/etc), the commit sub-process runs pi --model google-antigravity/... and fails without that auth.
Maybe instead of hardcoding models and providers, you could create a yaml config file for model/provider:
Create .memory/config.yaml (or .brain/config.yaml) for defaults:
committer:
model: google-antigravity/gemini-3-flash
thinking: low
memory_commit is currently hardwired to use Antigravity for its subagent, regardless of my active chat model.
Specifically:
COMMITTER_MODEL = "google-antigravity/gemini-3-flash"
model: google-antigravity/gemini-3-flash
So even if my main session has valid creds (Anthropic/OpenAI/etc), the commit sub-process runs pi --model google-antigravity/... and fails without that auth.
Maybe instead of hardcoding models and providers, you could create a yaml config file for model/provider:
Create .memory/config.yaml (or .brain/config.yaml) for defaults: