You read the OpenClaw docs on Dreaming — the feature that lets your agent promote short-term conversation memories into long-term persistent memory. It sounds powerful. You try to use it:

openclaw memory --help

No promote subcommand. You try:

openclaw memory promote "remember this"
Error: unknown command 'promote'

You're not doing it wrong. Dreaming is not available in the 2026.4.x release line. The official docs are ahead of the actual shipped build, and the team hasn't added a clear "not yet available" notice. This was confirmed in GitHub issue #61421.

Current status: Dreaming is described in the public docs as if it's available. It is not available in any 2026.4.x local build. The team has not given a ship date. This is a documentation-ahead-of-release situation, not a bug in your setup.

What "Dreaming" Is Supposed to Do

For context — here's what the feature is designed for, based on the docs:

It's a legitimately compelling feature. It's also not here yet.

What's Actually Available Right Now

Feature Available in 2026.4.x Notes
memory.md manual memory file ✓ Yes Works today — write facts directly into memory.md in your workspace
AGENTS.md / SOUL.md context injection ✓ Yes Injected into every turn — the practical equivalent of persistent memory
Workspace file-based memory ✓ Yes Agent can read/write any file in your workspace — use it for structured memory
Dreaming / openclaw memory promote ✗ Not Yet In docs, not in build — no ETA from the team
Automatic memory consolidation ✗ Not Yet Part of the Dreaming feature — same status
Memory search / retrieval ⏳ Partial Manual file-based search works; semantic search not yet shipped

How to Get Persistent Memory Right Now (Without Dreaming)

The good news: you don't need Dreaming to have a genuinely good memory system. Here's what we use in production setups:

The File-Based Memory Stack (Works Today)

memory.md — Your agent's persistent scratchpad. Tell your agent: "After every conversation where you learn something important about me, append it to memory.md." The file gets injected into your workspace context on every turn.

# In your workspace root
touch memory.md

# Instruct your agent to maintain it
# Example entry in SOUL.md:
# "Append key facts learned about Josh to memory.md after each session."

Structured Memory Logs

For higher-volume setups, create a memory/ directory with categorized files:

memory/
  facts.md       — persistent facts about you / your business
  decisions.md   — decisions made and reasoning
  projects.md    — project status and notes
  contacts.md    — key people and context

Instruct your agent to read and update the relevant file each session. This is essentially a manual version of what Dreaming is supposed to automate.

Heartbeat-Driven Memory Updates

If you're using heartbeats (scheduled agent runs), add a step in your HEARTBEAT.md that tells the agent to consolidate what it learned into memory files. This mimics the Dreaming loop manually.

# In HEARTBEAT.md:
## End of Cycle
- Review this conversation for new facts
- Append anything worth remembering to memory/facts.md
- Log the session summary to memory/heartbeat-log.md

When Will Dreaming Actually Ship?

Unknown. The GitHub issue was filed on April 5, 2026 and is actively being discussed. The team responded that Dreaming is on the roadmap but didn't give a specific release milestone.

Watch the GitHub releases page — when it ships, it'll be in the changelog. We'll also post here when it lands and how to actually use it.

Bottom line: Don't wait for Dreaming to build your memory system. The file-based approach above works well in production setups right now — and when Dreaming ships, migrating to it will be straightforward since it'll read the same files.