Probably the most absolute best ways to notice an AI assistant’s limits is this: it feels sharp at the moment, then oddly forgetful day after today.
You give an explanation for a workflow, indicate a call, or spend section an hour giving useful context. The chat goes smartly. Then the next session starts and also you may well be once more to repeating yourself.
That’s the disadvantage OpenClaw’s Dreaming characteristic is trying to unravel. It’s nevertheless marked as experimental at the time of writing, on the other hand the concept is already useful.
As an alternative of treating memory like a junk drawer, Dreaming periodically opinions what happened, helps to keep what problems, and leaves the rest at the back of. The target is understated: have the same opinion your AI keep in mind the proper problems.
Why “Dreaming”?
Because of sleep is when the thoughts sorts problems out.
We don’t keep in mind each tiny part from the day. We keep the useful parts, drop the noise, and each and every so regularly connect ideas one way or the other that most straightforward is sensible later.
OpenClaw is borrowing that idea.
When Dreaming runs, it opinions fresh conversations, appears to be for what helps to keep bobbing up, and promotes most straightforward the stronger indicators into longer-term memory. It moreover writes a plain-English summary of what it noticed.
That phase problems.
Numerous AI memory choices in point of fact really feel like black containers. Dreaming is more uncomplicated to consider on account of you’ll if truth be told take a look at what it decided was once price keeping.
How Dreaming Works
At a best stage, Dreaming runs on a schedule. The default is nightly at 3 AM, even if that can be customized.
The process is split into 3 levels.
1. Delicate Segment
That’s the cleanup pass.
Dreaming goes through fresh conversations, eliminates duplicates, and organizes the noise previous to the remaining gets promoted.
2. REM Segment
That’s the reflection pass.
The device appears to be for patterns, recurring topics, and problems that keep bobbing up. If you happen to keep talking concerning the an identical problem or downside, that starts to topic.
3. Deep Segment
That’s the position promotion choices happen.
The device scores what it came upon and most straightforward promotes probably the most tough recollections into long-term storage, very similar to MEMORY.md.
That phase problems on account of Dreaming is disposable by the use of default. It isn’t searching for to avoid wasting the entire thing perpetually. It is trying to stick what’s going to nevertheless be useful later.
The Dream Diary Is the Best possible Segment
After each cycle, OpenClaw writes a summary to DREAMS.md.
It’s basically a temporary diary get admission to in easy English showing what the device noticed. Something like:
"You've got been talking such a lot about group routing configurations lately. This seems like an ongoing problem."
That transparency makes the feature more uncomplicated to consider.
As an alternative of asking you to imagine a black box, OpenClaw presentations its artwork. You’ll be informed the diary, take a look at what it thinks is very important, and decide whether or not or now not the memory behavior feels good.
There is also a Wants UI you’ll open with Ctrl + I, which gives you a nicer method to browse that wisdom.
Recommendations on how one can Turn Dreaming On
Dreaming is disabled by the use of default, so you need to allow it explicitly.
You’ll do that in your OpenClaw config like this:
{
"plugins": {
"entries": {
"memory-core": {
"config": {
"dreaming": {
"enabled": true
}
}
}
}
}
}
You’ll moreover customize the schedule if you don’t want it operating most straightforward once a night. For example, chances are high that you’ll need a shorter length like each six hours, depending on how carefully you use your agent.
If you happen to would reasonably arrange it from chat, OpenClaw moreover is helping slash directions:
/dreaming on/dreaming off/dreaming status/dreaming have the same opinion
And if you want to have additional keep watch over, there are CLI apparatus for previewing what will also be promoted and understanding why something made the cut back.
That is useful should you’re the kind of one who likes seeing the scoring commonplace sense previous to trusting automation.
Why This Problems
Without some roughly memory consolidation, AI assistants usually have a tendency to turn into each forgetful or cluttered.
Forgetful is apparent: the assistant loses crucial context and becomes a lot much less useful over time.
Cluttered is the quieter downside. If a device saves quite a lot of, long-term memory fills up with noise. Earlier one-off details get began sitting beside exact preferences and lively projects. The result isn’t intelligence. It’s hoarding.
Dreaming goals for the middle ground: keep a lot much less, on the other hand keep the proper problems. If it in point of fact works smartly, your AI becomes additional useful over time because it helps to keep context additional cleanly.
Why This Lands at the Correct Time
This option moreover lands at a good time.
Numerous developers had been complaining lately that AI coding equipment in point of fact really feel a lot much less loyal than they used to. When that happens, memory problems additional. If an assistant already makes mistakes, it feels even worse when it moreover forgets your setup, preferences, and ongoing artwork – in particular in more hands-on setups like native AI help in VS Code.
Dreaming does not restore inclined reasoning. On the other hand it might in reality scale back one layer of friction by the use of helping the agent stay grounded throughout the context that if truth be told problems.
The post OpenClaw’s Dreaming Characteristic Is helping Your AI Keep in mind Higher gave the impression first on Hongkiat.
Supply: https://www.hongkiat.com/blog/openclaw-dreaming-feature/
0 Comments