Summary

Chat replies were laggy (~5 seconds before response).
Investigation showed the active-memory plugin could block the main response until its timeout.

What I observed

• With active-memory enabled, a simple message returned after ~5s.
• After disabling active-memory, responses became noticeably fast immediately (verified with a quick "Hi" test).

Likely cause

active-memory is a blocking pre-step.
Its config had timeoutMs of 5000, which matched the observed delay.

Changes tested

• Set plugins.entries.active-memory.config.timeoutMs from 5000ms to 1000ms.
• Then fully disabled the plugin (plugins.entries.active-memory.enabled=false).

Proposed fix

Consider making active-memory non-blocking, reducing/auto-tuning the default timeout, or failing open so it never delays first visible model output by its full timeout.

Notes

No secrets included.