openclaw - 💡(How to fix) Fix [Bug]: Make Dream Diary narrative timeout configurable for slow/serial local model backends [1 comments, 2 participants]

Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

Dreaming’s Dream Diary narrative subagent wait timeout appears to be hardcoded to 60 seconds. On local model backends such as LM Studio, prompt processing may be serialized even when OpenClaw starts multiple dreaming narrative runs in parallel/queued form. This can cause dreaming narrative runs to exceed 60s and fail even though the backend is still working normally.

Steps to reproduce

Dreaming starts multiple narrative prompts around the same sweep. LM Studio processes prompt work serially, so queued narrative runs can exceed the 60s wait budget.

Related logs include errors like:

EmbeddedAttemptSessionTakeoverError: session file changed while embedded prompt lock was released and/or timeout-related narrative/embedded-run logs if present.

Expected behavior

The Dream Diary narrative timeout should be configurable, or OpenClaw should reduce narrative concurrency for local/serial model backends so dreaming can complete reliably.

Actual behavior

Related logs include errors like:

EmbeddedAttemptSessionTakeoverError: session file changed while embedded prompt lock was released and/or timeout-related narrative/embedded-run logs if present.

OpenClaw version

2025.5.12

Operating system

Ubuntu

Install method

npn

Model

qwen/qwen3.6 35B

Provider / routing chain

openclaw -> local

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

Add a user-facing config key for the Dream Diary narrative timeout, for example under: json { "plugins": { "entries": { "memory-core": { "config": { "dreaming": { "narrative": { "timeoutMs": 300000, "concurrency": 1 } } } } } } } Exact config shape can differ, but the important part is making the current 60s timeout tunable.