AI With Perfect Memory? Here's How (OpenClaw True-Recall Base) LOCAL and FREE

Okay, so here we go. So, finally I was able to get this completed. Um, I went back and forth trying to get it right so many times and so many like

reiterations. And as I'd mentioned in

reiterations. And as I'd mentioned in other videos, probably multiple times, how when you tell the AI to do something, it'll say it did it and

it'll check it and then later find out it didn't actually do it and or andor there's a problem, right? So anyways,

this is true recall. Okay, so um it's basically Jarvis memory but like way better, right? And so I decided to

make it into like multiple uh tiers, right? So I I after Jarvis memory and I found out that I could I could write directly to Quadrant, I got

rid of Reddus and this only only needs quadrant now or or or you could use a an online vector

database. I mean so the options are out

database. I mean so the options are out or whatever you want.

And so I I what I had done was I was trying to try just make like this gems model like V2 and that was a combination of the base and gems, right? Then I I

got to thinking. I was like, you know, I really use the base a lot, right? Like

the gems is still needs a little work.

And then I had I came up with blocks.

And I'll explain that in a second. But

basically, the base is probably the most handy thing. And I figured that these

handy thing. And I figured that these these could be like add-ons. So that way if you just want the base, it it does I mean it almost does

everything I want, right? Because I just I'll show you in a second. But it it's really pretty comprehensive and it's simplified, right? So it's really

simplified, right? So it's really simplified. Anyways, my point is you

simplified. Anyways, my point is you have gems and you have blocks or things that I'm going to come out with later that add on to this or you don't have to add on to it. Like it's all already

complete. Okay. But if you want to do

complete. Okay. But if you want to do something a little bit more, I am going to work on these and then hopefully I can get off this memory kick and like start doing videos on other things, you

know. Um, anyways, so I I had it I

know. Um, anyways, so I I had it I created a quick quick install and then it it'll give you like it'll prompt you with like the questions like uh what's

your your Quadrant IP address, your Olama IP address, things like that. And

then it'll ask you if you want to proceed.

So, it's pretty simple installation or there's like manual steps that you could just do it yourself.

Um, let's see.

Oh, requirements. I tried to get this as

requirements. I tried to get this as thorough as possible. So, realistically,

you could not use the GitHub and still be able to install it yourself without just with the instructions. That was

kind of my goal. Okay. So, the

architecture is, you know, you chat with the AI and then the session gets recorded in openclaw here and the watcher watches for this to close. So

it's it calls it turns, right? So you

chat, it responds one turn. This Damon

sees it and then it pops it in. It takes

it it strips out all the extra stuff because it's it it saves it with all the special characters and the thinking of the model and all that stuff. Well, this

one strips all that out and then and uses Olama embedding model to put it into quadrant memories. So, I mean, it's

pretty cool. I mean, it's actually very

pretty cool. I mean, it's actually very simplified. Um, I'm I'm really glad that

simplified. Um, I'm I'm really glad that I went this route because it it it keeps it simple for, you know, just just basic like portable memory, right? Like cuz

whenever I if I move to another project or something like if a better open claw comes around and I want to use the memory, having this simple kind of

design works really nice. So anyways,

uh, message me. So you could read through. I'm not going to go through

through. I'm not going to go through this entire thing, but I tried to include as much detail as possible and the way that it's designed like like anything I could think of, I tried to

tried to get in here so that way it was very clear like how it works, right? So,

every detail uh mostly because if you want to do it yourself and you're like, I'm not going to use a GitHub thing or whatever, you just want to recreate it, you could probably grab text from here and plug it

into your AI and get the same thing. So,

you don't have to use the GitHub. It's

just uh someone had recommended it and I was like, "Oh, yeah, that's actually a pretty good idea." Because I already had a GitHub account. Oh, and in the other video I mentioned, you know, my GitHub

account is um is blocked uh for whatever reason. It's still not open.

They said that it could take like uh up to like 40 days, some ridiculous thing. So, I put in the ticket, you

thing. So, I put in the ticket, you know, four days ago. I complained that they weren't doing anything. I keep

sending them messages and they just don't respond. And according to online,

don't respond. And according to online, that's just how they roll. Anyways, so I have this on my local um get giddy g get g get g get g get g get g get g get g get g get g get gy I don't

I'm not sure how it's said but and then I have it on GitHub so you can I'm sorry uh what is it gitlab because I I wanted

to have some other official place so that everyone could know you know all the codes out there there's no hidden you know anything in it right so anyways now I was going to give you a little bit

of an example So this is kind of how it works, right? So I um I create a new

works, right? So I um I create a new session. Now this is just

session. Now this is just uh sorry, where did I create the new session? So there I just did the latest

session? So there I just did the latest update and it changed some things around where like when I create a new session, it doesn't dump the text from before.

Not sure if that's a bug or a feature.

Right. So I start a new session. And

this is just kind of an example of me kind of probing it to find out like how it does what it's doing with the memories. Right? So I said when I say do

memories. Right? So I said when I say do you remember that where where do you search? And so when you ask if I

search? And so when you ask if I remember something I search Q which that's what I abbreviated to in her memory. So that that it's always the

memory. So that that it's always the quadrant memories. Okay. So it said she

quadrant memories. Okay. So it said she said I select I search that first.

That's where all our confrontation are captured in real time. If Q doesn't turn up results, then she checks her built-in memory that the way that open call normally works. So, this doesn't break

normally works. So, this doesn't break openclaw functionality. And I was like,

openclaw functionality. And I was like, awesome. That's exactly what I wanted.

awesome. That's exactly what I wanted.

And she said, glad to hear it on the memory front. What else? I said,

uh, so I did notice. So, so I tried to turn off injection, right? So, so here's the here's part of the cool part. I

think there's some debugging still on because it gets all printed out here in the in the console. Right. So, uh here is

okay. There's two different things.

okay. There's two different things.

Agent fires is where it's capturing and then it it has like a thing that says it parsed it and there were four exchanges and it inserted it into it got embedded.

Right? So, I think that's this section right here. So it keeps doing that

right here. So it keeps doing that because every time we talk back and forth, it grabs it. Now this is the injection portion for the gems, right?

So it says the search return zero results threshold. I have it set at 08.

results threshold. I have it set at 08.

So it has to have a a very high match to inject anything. Uh and so didn't meet

inject anything. Uh and so didn't meet threshold. So skipping injection. So

threshold. So skipping injection. So

again, and you could see every time I respond or I respond to it, it checks the threshold for results from the gyms. But anyways, that's not what this video

is about. That's just still in there.

is about. That's just still in there.

But but the conversation I was having with Kimmy was I thought I turned it off. And so she said she didn't turn it

off. And so she said she didn't turn it off. It's still active. And then the job

off. It's still active. And then the job to create the gems. And I said, "I wanted you to turn them both off. I think you said before you

both off. I think you said before you paused them already." And she's like, "You're right. Let me check what we have

"You're right. Let me check what we have actually have discussed. Searching my

memories now." Which is supposed to be Q, right? Um and then I think in the new

Q, right? Um and then I think in the new version, she didn't ever respond. So I

said I had to say, "What memories did you search?" And she said, "You're

you search?" And she said, "You're catching me. I actually didn't search Q

catching me. I actually didn't search Q my memories. I jump straight to checking

my memories. I jump straight to checking the collection state with curl commands instead of doing a proper semantic search like I should have which I think this is just the Kimmy model is kind of like weird like this anyways she said

let me do it right now searching my memories for our conversation and pausing turning off injection and so again she didn't come back and respond I

said what did you find and so she searched Q memories and she she didn't see which she's right I think that I was mistaken But she did find. So, one of the things

I did tune up that that I did actually about a couple hours ago was I I didn't have the memory sorted by date. I had

the date timestamp, but I I just assumed that she was going to take it into consideration, but she wasn't. So, what

I did was I I had the the the job that searches the memories, the script to to order them by the date time. So that way I was thinking like if I change my mind

on something, I want the most recent conversation. I don't want you to spit

conversation. I don't want you to spit out an old conversation, right? So So

this is what she found. Uh I paused the t the timer curator job not injection.

You ask how does it get deter? I mean

she quoted me how does it get determined? What I what gets injected? I

determined? What I what gets injected? I

reported context injection at 05 threshold. No record of pause injection.

threshold. No record of pause injection.

Did you ask somewhere else or should I dis disable it now? But but what's cool is that she she decided she automatically searched the memories, right? So this is all from the memories.

right? So this is all from the memories.

This is not from the built-in. But

that's kind of the gist of it. If anyone

has any questions, thanks for watching and uh I hope that you find it useful and you could use it if you choose to.

Thanks. Take care.