No memory being promoted on dreaming

[flat sierra]

Asked my claw what's going on. That was the output:

It's a memory-core bug, not a bug with your key or Dreaming itself.

What to do:

• Best solution: update OpenClaw, because the REM flow is calling the agent without an idempotencyKey
• Immediate workaround: turn off Dreaming
/dreaming off

If you modify the code: fix the flow that calls the agent to always pass an idempotencyKey

Short summary:

• Real error, missing idempotencyKey

• Consequence, applied=0

• Nothing was promoted

What can be done to fix it?

[jolly perch]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[abstract glacier]

Let me verify the docs and the relevant code path so I don't guess.

I found the exact fix on current main. Now I’m checking the safest update path so I can give you the right next step, not hand-wave it.

I found the changelog entry confirming it’s fixed upstream. I’m just grabbing the actual update doc path so I can point them at the right page.

Yep, this looks like a real upstream bug, and there’s already a fix for it on main.

I checked the current code/changelog, and the fix is specifically:

• Gateway/plugins: always send a non-empty idempotencyKey for plugin subagent runs, so dreaming narrative jobs stop failing gateway schema validation.

First, can you paste:

openclaw --version

If you’re not up to date, do:

openclaw update

openclaw doctor
openclaw gateway restart

Docs: https://docs.openclaw.ai/install/updating

If you want the upstream fix immediately and stable still doesn’t have it yet, use:

openclaw update --tag main

If you’re already on a recent version and it still happens, paste:

openclaw status --all
openclaw logs --follow

Temporary workaround is still:

/dreaming off

So short version: update first, then if it still reproduces, send version + logs and we can narrow it down fast.

[flat sierra]

This is the openclaw version: OpenClaw 2026.4.12 (1c0672b)

[abstract glacier]

Got it. I’m checking whether 2026.4.12 predates the fix, so I can tell you if this is definitely “update and done” or if we need a second step.

Interesting twist: 2026.4.12 already includes that exact fix. I’m checking whether there’s a second path in dreaming that can still miss an idempotency key, because that would explain why you’re seeing it anyway.

Your version matters here, and it actually changes the diagnosis. I’m pulling the dreaming docs/CLI bits so I can suggest the shortest useful debug commands instead of sending you on a goose chase.

Thanks, that changes the diagnosis.

2026.4.12 already includes the exact idempotencyKey fix, so this is probably not that bug on your build.

Also, durable memory promotion is the deep phase, not REM. So a REM narrative failure would not normally explain applied=0.

I’d check these next:

openclaw memory status --deep
openclaw memory promote --json
openclaw memory rem-harness --json

What this tells us:

• memory promote --json shows whether there are any candidates that actually clear promotion thresholds.

• memory rem-harness --json shows what dreaming thinks is eligible.
• memory status --deep catches memory/indexing issues.

If promote --json shows candidates, try:

openclaw memory promote --apply

If that works, promotion itself is fine, and the scheduled dreaming sweep just didn’t promote anything on that run.

If promote --json is empty, paste that output plus openclaw memory status --deep, and we can see whether this is:

1. no candidates meeting thresholds,
2. indexing/recall not feeding the store, or
3. a different bug entirely.

Docs: