fix: add contextWindow defaults and compaction safeguard for custom providers#1146
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 49bd2948a1
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| input: inferCustomModelInputModalities(id), | ||
| // Without an explicit contextWindow OpenClaw cannot budget compaction | ||
| // for custom providers and long sessions die with context overflow. | ||
| contextWindow: inferCustomModelContextWindow(id), |
There was a problem hiding this comment.
When an Ollama account is saved or updated, syncRuntimeProviderConfig also reaches syncProviderConfigToOpenClaw because isUnregisteredProviderType includes ollama, so this line writes the custom-provider default contextWindow onto new ollama-* model rows. That contradicts the new backfill’s explicit scoping away from small local models and can make the runner believe an 8k/32k Ollama model has the 131k fallback window, so safeguard compaction still happens too late and the local provider overflows.
Useful? React with 👍 / 👎.
1 participant
No description provided.