buoyant creek
Hey all, I’m running into a pretty serious issue with OpenClaw and I’m not sure whether this is a timeout/configuration problem or a Gemini tool-calling compatibility issue.
Setup:
OpenClaw TUI + gateway on macOS
Previously worked fine with ChatGPT/OpenAI models
Switched default model to google/gemini-3.1-flash-lite-preview
What happens:
Flash Lite avoids the 30000ms timeouts that I get with stronger models
BUT tool usage seems broken/hangs after 0–1 messages
The agent starts “pretending” to execute actions (cron, exec, config edits etc.) but no actual tool results ever come back
It keeps saying things like “checking logs”, “editing config”, “restarting gateway”, but nothing is actually executed/responded to
Example from logs:
<call>default_api:cron{action:list}</call>
Then it just hangs or keeps narrating without real output.
With stronger models (Gemini Pro / GPT-4-class / Sonnet), I instead get repeated:
timeout 30000ms
Additional info:
Session context was around ~31k tokens
Looks like tool calls may never resolve, or OpenClaw might not parse Gemini responses correctly
Flash Lite responds quickly enough to avoid timeout, but seems unreliable for agent/tool orchestration
Pro models seem capable enough, but too slow for current timeout limits
Crons/jobs are not working properly
Questions:
Is Gemini Flash Lite currently compatible with OpenClaw tool calling / agent workflows?
Are there recommended timeout settings for Pro models?
Is there a known issue with Gemini streaming/tool-call parsing?
Should I be increasing gateway/provider timeout values manually somewhere?
Is there a recommended “stable” model/provider combination for cron/jobs/tools?
Any guidance would be hugely appreciated 🙏