Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

Bug type

Behavior bug (incorrect output/state without crash)

Summary

Active Memory does trigger in direct-chat sessions, but it consistently stalls for about 30 seconds and then times out, visibly blocking the reply.

I tested two configurations:

1. Explicit plugin model:

• plugins.entries.active-memory.config.model = "nvidia/stepfun-ai/step-3.5-flash"

2. Follow resolved/default session model:

• removed plugins.entries.active-memory.config.model
• set plugins.entries.active-memory.config.modelFallbackPolicy = "resolved-only"

In the second case, runtime logs showed the provider as 39api/gpt-5.4.

Both configurations behaved the same: Active Memory started, consumed ~30s, and timed out.

Steps to reproduce

1. Enable Active Memory in a direct chat session.
2. Set:

• queryMode = "message"
• timeoutMs = 30000

3. Test once with explicit model:

• model = "nvidia/stepfun-ai/step-3.5-flash"

4. Restart gateway and send a normal message.
5. Observe timeout.
6. Remove model, set:

• modelFallbackPolicy = "resolved-only"

7. Restart gateway and send another normal message.
8. Observe the same timeout again.

Expected behavior

Active Memory should either complete quickly enough for interactive chat or fail open gracefully without blocking the session for ~30 seconds.

Actual behavior

Direct-chat turns are delayed by ~30 seconds, then Active Memory times out.

OpenClaw version

2026.4.11

Operating system

Linux

Logs, screenshots, and evidence

[plugins] active-memory: agent=main session=agent:main:main start timeoutMs=30000 queryChars=336
[agent/embedded] embedded run failover decision: ... reason=timeout provider=39api/gpt-5.4
[diagnostic] lane task error: ... error="FailoverError: LLM request timed out."
[plugins] active-memory: agent=main session=agent:main:main done status=timeout elapsedMs=30158 summaryChars=0

Earlier run with explicit model also timed out:

[plugins] active-memory: agent=main session=agent:main:main start timeoutMs=30000 queryChars=334
[agent/embedded] embedded run failover decision: ... reason=timeout provider=nvidia/stepfun-ai/step-3.5-flash
[diagnostic] lane task error: ... error="FailoverError: LLM request timed out."
[plugins] active-memory: agent=main session=agent:main:main done status=timeout elapsedMs=30062 summaryChars=0

Steps to reproduce

NOT_ENOUGH_INFO

Expected behavior

NOT_ENOUGH_INFO

Actual behavior

NOT_ENOUGH_INFO

OpenClaw version

2026.4.11

Operating system

Debian13

Install method

No response

Model

gpt-5.4/nvidia/stepfun-ai/step-3.5-flash

Provider / routing chain

openclaw -->Gateway --> gpt-5.4/nvidia/stepfun-ai/step-3.5-flash

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

Additional information

This does not look like “Active Memory never triggers.” It does trigger, but the embedded path appears to stall and hit timeout across multiple model configurations.

Temporary mitigation: disable Active Memory.