A developer encountered a persistent issue with the OpenClaw gateway where model changes were not taking effect, leading to timeouts. The problem stemmed from a configuration conflict where the `modelPolicy.allow` list, acting as a hard allowlist, overrode the specified primary model. When the primary model was not on this list, OpenClaw silently selected a dead model from the allowlist instead of failing or warning the user. The resolution involved adding the new primary model to the allowlist and using `grep` to find the overriding configuration, as standard checks reported the correct primary model while a different one was being dispatched. AI
IMPACT Highlights potential pitfalls in LLM gateway configuration and model selection logic.
RANK_REASON Developer troubleshooting a specific configuration issue with an open-source gateway.
- custom-groq/openai/gpt-oss-120b
- custom-integrate-api-nvidia-com/moonshotai/kimi-k3
- Groq
- moonshotai/Kimi-K3
- Nvidia
- OpenClaw
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →