💡 If you have been keeping daily notes in OpenClaw for months or years, here is some news that should make you genuinely excited: OpenClaw v2026.4.9 ships a fully grounded REM backfill lane that can replay your entire note history into the Dreams memory system. Everything you wrote before Dreams was active? It can now become durable, queryable memory.
Let me explain what that means and how to use it.
Background: How Dreams Works
OpenClaw's Memory/Dreaming subsystem — called Dreams — is the layer that turns transient session notes into long-term structured memory. During a normal session, observations, decisions, and facts flow into short-term memory. Periodically, the Dreams process runs and promotes the most valuable short-term memories into "grounded facts" — durable, semantically indexed records that persist across sessions and can be retrieved by future context windows.
Think of it like sleep. During the day, your brain accumulates experiences. At night, REM sleep consolidates the meaningful ones into long-term memory while discarding the noise. Dreams does the same thing for your OpenClaw sessions.
The problem, until now, was that Dreams only worked on notes generated after you enabled it. If you had two years of daily notes sitting in your workspace from before you started using the Dreams feature, those were invisible to the memory system. A lot of potentially valuable context was locked out.
REM backfill changes that.
What REM Backfill Does
The grounded REM backfill lane is a pipeline that takes historical notes as input and runs them through the same consolidation process that Dreams uses for live sessions. The key components:
rem-harness --path — The entry point. Point it at a directory or file path containing historical notes, and the harness orchestrates the full backfill run. It handles batching, ordering, and fault tolerance.
Durable-fact extraction — Each note is analyzed for facts worth keeping: decisions made, preferences expressed, project context, people mentioned, technical details logged. Extracted facts are written to the grounded memory store in the same format as live-session Dreams output, so they are immediately available for retrieval.
Live short-term promotion integration — The backfill lane integrates with the live short-term memory pipeline. Facts extracted from historical notes participate in the same promotion logic as fresh memories — they can surface in context windows, influence suggestions, and be referenced by future Dreams runs.
Diary commit/reset flows — Backfill is designed to be transactional. After each batch, the diary state is committed. If a run fails partway through, you can reset to the last committed state and resume without re-processing already-completed notes. This makes it safe to run backfill on large note archives without worrying about partial runs leaving the memory store in an inconsistent state.
The New Control UI
Dreams management in the Control UI has also received a significant upgrade alongside the backfill feature.
There is now a structured diary view with timeline navigation. You can browse your diary entries chronologically, see which ones have been processed by Dreams, and identify gaps or entries that failed to process cleanly.
Backfill and reset controls are available directly in the UI. You no longer need to reach for the CLI to kick off a backfill run or roll back a problematic one. The UI also shows progress for in-flight backfill operations.
Traceable dreaming summaries let you inspect what happened during each Dreams run — which source entries were processed, which facts were extracted, and what the confidence levels were. This is invaluable for understanding why Dreams surfaced a particular memory.
The Scene lane now shows promotion hints alongside each short-term memory, giving you a real-time view of which items are candidates for promotion to grounded facts on the next Dreams cycle.
Finally, there is a safe clear-grounded action for when you need to reset your grounded memory store entirely. It requires explicit confirmation and is intentionally designed to be hard to trigger accidentally.
How to Use REM Backfill
The basic workflow is straightforward:
# Point the harness at your historical notes directory
rem-harness --path ~/notes/daily
# Or target a specific date range
rem-harness --path ~/notes/daily/2025
# Check status after a run
rem-harness --status
The harness will process notes in chronological order, extract facts, and commit to the diary store incrementally. For a large archive, expect the first backfill run to take a while — fact extraction is not cheap, and the harness deliberately paces itself to avoid hammering the model API.
If you interrupt a run or it fails, resume cleanly with:
rem-harness --path ~/notes/daily --resume
This picks up from the last committed checkpoint.
After the backfill completes, open the Control UI diary view to inspect the results. You should see your historical entries appearing in the timeline with extracted facts attached.
What Gets Extracted?
The extraction quality depends on how you write your notes, but the system is broadly good at pulling out:
- Technical decisions and their rationale ("chose PostgreSQL over MySQL because of JSON support")
- Project context and status ("MySmartTour v2 launch targeting Q2")
- Preferences and patterns ("prefers morning sessions for deep work")
- People and relationships referenced in notes
- Recurring topics and themes that suggest long-term priorities
Facts are stored with source provenance, so you can always trace a grounded memory back to the original note entry that produced it.
Hardened Inputs in This Release
Alongside the new feature, v2026.4.9 also hardens the backfill pipeline against malformed inputs. Diary headings are now normalized before processing (fixing edge cases where non-standard heading formats caused extraction failures on older notes), and diary write operations are more robust against concurrent access.
If you have older notes with inconsistent formatting, the backfill lane should handle them more gracefully than prior versions of the Dreams engine would have.
The Bottom Line
REM backfill is one of those features that makes an existing capability retroactively more valuable. If you have months of daily notes sitting in your workspace, you now have a path to make all of that context available to Dreams — and by extension, to every future session that benefits from grounded memory retrieval.
Run the backfill. Let Dreams sleep on your history. See what surfaces.