feat: context compaction never triggers mid-turn, so long tool-calling runs die on context limit

Check Existing Issues

• I have searched all existing open AND closed issues and discussions and found none comparable to my request.

Verify Feature Scope

• I believe this feature request is appropriately scoped for the Issues section as described above.

Problem Description

My Use Case: When I give the chat a long, multi-step task (e.g., "Analyze this data file one by one and generate a report"), the context window often runs out in the middle of processing.

The Current Behavior: The chat completely stops mid-task. I have to manually click "Continue Chat," which finally triggers the context compaction feature and allows the LLM to resume.

The Expected/Agentic Behavior: Other agentic platforms handle this seamlessly. They automatically trigger context compaction during tool calls or mid-generation the moment the limit is hit, without requiring human intervention to keep the loop going.

Proposed Solution

Allow context compact trigger during tools calling, not only limited to after chat

Alternatives Considered

No response

Additional Context

No response