1 00:00:00,200 --> 00:00:01,240 And week three. 2 00:00:01,280 --> 00:00:02,680 Day four is a go. 3 00:00:02,720 --> 00:00:03,880 Let's get started. 4 00:00:03,920 --> 00:00:08,000 So last time we did a stock picker and we just have a little tiny bit more to put into that. 5 00:00:08,000 --> 00:00:10,560 And then we're going to move on with our next project. 6 00:00:10,600 --> 00:00:12,320 A developer agent. 7 00:00:12,600 --> 00:00:14,640 But first another repetition. 8 00:00:14,640 --> 00:00:17,960 I hate to repeat these things, but sometimes it's good to drum it in. 9 00:00:18,240 --> 00:00:20,760 Building a crew project involves five things. 10 00:00:20,800 --> 00:00:23,120 First of all, crew I create crew. 11 00:00:23,560 --> 00:00:28,440 The name of the project to set up your crew, build those directories and those files. 12 00:00:28,480 --> 00:00:33,760 Number two, you find the YAML files for your agents and tasks, and you fill them in to define your 13 00:00:33,760 --> 00:00:35,040 agents and tasks. 14 00:00:35,280 --> 00:00:40,400 Number three, you go to the crew.py module, which is where you actually create the instances. 15 00:00:40,400 --> 00:00:44,960 And you use decorators to identify the agents and tasks that you'll be using. 16 00:00:45,000 --> 00:00:47,320 And then you create your crew itself. 17 00:00:47,360 --> 00:00:53,240 And this is where you can have structured outputs to make sure that the outputs conform to a schema. 18 00:00:53,240 --> 00:00:58,960 And you can use tools, both tools like Serpa, that crew provides for us that run remotely, and then 19 00:00:58,960 --> 00:01:03,060 custom tools that we build locally, like the thing that sends a push notification. 20 00:01:03,340 --> 00:01:10,020 And then number four, you update Main.py to set any inputs so that we can pass something in, configure 21 00:01:10,020 --> 00:01:12,140 the fields that are templated with the curlies. 22 00:01:12,500 --> 00:01:15,100 And finally we run with crew. 23 00:01:15,140 --> 00:01:18,220 I run and off goes our project. 24 00:01:18,740 --> 00:01:22,420 So I now want to cover a feature of crew called memory. 25 00:01:22,420 --> 00:01:28,060 And this is a feature that is a bit more prescriptive, a bit more opinionated in the crew framework. 26 00:01:28,100 --> 00:01:34,140 Memory, of course, is talking about how you provide information, contextual information to Llms each 27 00:01:34,140 --> 00:01:39,180 time you call them, and you can implement that yourself just by storing variables and then passing 28 00:01:39,180 --> 00:01:43,660 them in when you do things like creating tasks so you can do it the sort of manual way. 29 00:01:43,820 --> 00:01:49,180 But the crew framework also comes with some building blocks that lets you use their constructs around 30 00:01:49,180 --> 00:01:52,140 memory out of the box, and that comes with pros and cons. 31 00:01:52,180 --> 00:01:57,020 The pro is that you get up and running quickly, and you can use a lot of the thinking that they put 32 00:01:57,020 --> 00:01:57,820 behind this. 33 00:01:58,020 --> 00:02:03,610 The con is that there's there's a learning curve, and it obscures some of the detail of how prompts 34 00:02:03,650 --> 00:02:05,250 actually work behind the scenes. 35 00:02:05,250 --> 00:02:10,730 So as as with many times when you're adopting a framework like this, it's something to be aware of 36 00:02:10,810 --> 00:02:13,050 the benefits and the trade offs of doing so. 37 00:02:13,410 --> 00:02:18,130 But let's say that we are going to embrace cruise way of handling memory and talk about what it actually 38 00:02:18,130 --> 00:02:18,610 does. 39 00:02:18,850 --> 00:02:24,330 Well, it has five different types of memory, five different frameworks that you can include. 40 00:02:24,530 --> 00:02:26,850 And one of them is called short term memory. 41 00:02:26,850 --> 00:02:34,570 And this is just about storing recent interactions using a vector database in in a in a rag way. 42 00:02:34,570 --> 00:02:37,210 If you're familiar with retrieval augmented generation. 43 00:02:37,210 --> 00:02:40,650 And you don't need to be for this course because we're just going to put the code in there and see it 44 00:02:40,650 --> 00:02:41,130 run. 45 00:02:41,130 --> 00:02:43,250 But if you do no, then this will make more sense. 46 00:02:43,250 --> 00:02:49,850 So this will allow agents to access recent relevant information when they are currently executing. 47 00:02:50,450 --> 00:02:56,330 And then a different concept called long term memory is when more important information is stored in 48 00:02:56,330 --> 00:03:03,010 a SQL database for longer term recall to build up knowledge over over a longer period of time. 49 00:03:03,610 --> 00:03:08,510 And then there's something called entity memory that's very similar to short term memory, actually. 50 00:03:08,550 --> 00:03:15,310 It's it's basically when there's things about people, places and concepts, then those can also be 51 00:03:15,310 --> 00:03:22,430 stored in a Rag database for vector based similarity search and to be included in the context. 52 00:03:22,830 --> 00:03:26,110 And then there's crew describes this as another kind of memory. 53 00:03:26,110 --> 00:03:27,230 But I think it's a bit misleading. 54 00:03:27,230 --> 00:03:33,990 I think what they call contextual memory is just a sort of umbrella term for the short term, long term 55 00:03:33,990 --> 00:03:42,070 and entity memory that can all together be queried and passed in as context when prompting an LM and 56 00:03:42,070 --> 00:03:43,910 crew abstracts all this away from you. 57 00:03:43,910 --> 00:03:48,190 So it's just going to be a few lines of code to have all of these types of memories running. 58 00:03:48,390 --> 00:03:50,430 But in doing so, as I say, big benefit. 59 00:03:50,470 --> 00:03:53,070 A lot of work happens behind the scenes. 60 00:03:53,070 --> 00:03:57,030 And perhaps also there the trade off is that you've got less visibility into it. 61 00:03:57,030 --> 00:04:01,270 So if things don't go the way you want, then it's a bit harder to debug and figure out what's going 62 00:04:01,270 --> 00:04:01,630 on. 63 00:04:01,910 --> 00:04:08,500 And then there is another kind of memory called user memory, which is to store user specific information. 64 00:04:08,500 --> 00:04:14,940 And actually, at least as of now in Crewe, this is a concept that they support and have some frameworks 65 00:04:14,940 --> 00:04:21,300 around, but it's mostly left up to you to be querying user memory and then inserting it into the prompt 66 00:04:21,300 --> 00:04:22,900 or providing it at the right time. 67 00:04:22,900 --> 00:04:28,020 So user memory is a bit of an odd one out here, and I suspect that they're looking to build more into 68 00:04:28,020 --> 00:04:28,940 that in time. 69 00:04:29,260 --> 00:04:33,180 And for now, for the code that we're about to look at, we're just going to really look at contextual 70 00:04:33,180 --> 00:04:33,540 memory. 71 00:04:33,540 --> 00:04:40,300 So short term long term and entity memory and seeing how we can incorporate that into our stock picker 72 00:04:40,300 --> 00:04:40,980 solution. 73 00:04:41,420 --> 00:04:44,540 And so we're back in the stock picker project in cursor. 74 00:04:44,540 --> 00:04:47,260 And we are looking at the crew.py module. 75 00:04:47,580 --> 00:04:53,100 And I'm going to start by putting in some new imports in here, which are interesting ones from crime 76 00:04:53,100 --> 00:04:53,460 memory. 77 00:04:53,460 --> 00:04:58,500 We're going to import long term memory short term memory and entity memory. 78 00:04:58,500 --> 00:05:00,180 The three types we'll be working with. 79 00:05:00,220 --> 00:05:03,660 And I do believe you can also have user memory in there too. 80 00:05:03,700 --> 00:05:06,220 But then you have to manually manage it yourself. 81 00:05:06,800 --> 00:05:13,320 And then from my memory storage rag storage, we're importing a class called Rag storage for vector 82 00:05:13,320 --> 00:05:14,880 based retrieval. 83 00:05:15,160 --> 00:05:24,120 And with that we're also going to import from long term memory SQLite storage, long term memory SQLite 84 00:05:24,160 --> 00:05:26,720 storage object like that a class. 85 00:05:26,920 --> 00:05:27,400 All right. 86 00:05:27,400 --> 00:05:30,640 So that's a few things that we're now going to put to good use. 87 00:05:30,640 --> 00:05:37,080 We're now going to go to the crew function the function that creates the crew within this this module. 88 00:05:37,080 --> 00:05:39,320 And you can see where we made our manager. 89 00:05:39,320 --> 00:05:41,320 And we've got a few more things to make. 90 00:05:41,360 --> 00:05:48,120 As as cursor is trying to prompt us, they're trying to ignore cursors, insisting let's do it ourselves. 91 00:05:48,320 --> 00:05:48,720 All right. 92 00:05:48,720 --> 00:05:55,960 So we are going to to want to be creating a short term memory a long term memory and an entity memory. 93 00:05:55,960 --> 00:05:58,280 And we're going to do them one by one. 94 00:05:58,280 --> 00:06:04,440 So let's start by saying that the short term memory, which is the one that we'll begin with, short 95 00:06:04,480 --> 00:06:08,780 term memory, trying not to press a tab or it's going to fill it all in for me. 96 00:06:08,780 --> 00:06:10,820 But I'm going to I'm going to do it right here. 97 00:06:10,860 --> 00:06:14,940 Short term memory is going to be something which has rag storage. 98 00:06:15,340 --> 00:06:22,460 We we come up with a provider which is OpenAI, and a model, an embedding model to generate vectors 99 00:06:22,460 --> 00:06:23,420 from text. 100 00:06:23,420 --> 00:06:24,620 And we'll be using this one. 101 00:06:24,620 --> 00:06:27,860 And you can substitute in whatever models you would like here. 102 00:06:27,860 --> 00:06:29,220 It's going to be short term. 103 00:06:29,220 --> 00:06:35,380 And we give it a path to where we'd like it to create that memory, that vector store as memory. 104 00:06:35,620 --> 00:06:41,540 And it will use chroma as it happens, something which people who've taken my, uh, alarm engineering 105 00:06:41,540 --> 00:06:42,540 course know very well. 106 00:06:42,540 --> 00:06:43,460 I love chroma. 107 00:06:43,700 --> 00:06:47,180 All right, so that's the short term memory. 108 00:06:47,340 --> 00:06:50,660 Let's also create some long term memory. 109 00:06:50,660 --> 00:06:51,620 So here it is. 110 00:06:51,620 --> 00:06:57,260 The long term memory is going to be just simply creating an instance of this of this class long term 111 00:06:57,260 --> 00:06:58,820 memory SQLite storage. 112 00:06:58,820 --> 00:07:01,020 And we'll give that also a place to go. 113 00:07:01,060 --> 00:07:04,660 We'll make a database file also in the same directory. 114 00:07:04,980 --> 00:07:07,140 So that's the long term memory object. 115 00:07:07,500 --> 00:07:13,850 And now finally, thirdly, we're going to create a, uh, entity memory object. 116 00:07:13,850 --> 00:07:15,810 So entity memory. 117 00:07:16,290 --> 00:07:17,250 And there it is. 118 00:07:17,450 --> 00:07:19,450 So oh I see. 119 00:07:19,490 --> 00:07:19,730 Hang on. 120 00:07:19,770 --> 00:07:20,570 We've got two of them. 121 00:07:20,570 --> 00:07:21,530 Let's do that. 122 00:07:21,930 --> 00:07:22,810 Looks like that's good. 123 00:07:22,850 --> 00:07:25,250 So entity memory is going to be an entity memory object. 124 00:07:25,250 --> 00:07:28,930 It's also going to be a rag storage object. 125 00:07:28,930 --> 00:07:33,770 We give the provider and the embeddings model and we put it in the memory folder. 126 00:07:33,970 --> 00:07:40,370 So here we have our three types our short term memory our long term memory and our entity memory. 127 00:07:40,530 --> 00:07:44,530 And now we get to the place where we create our crew. 128 00:07:44,850 --> 00:07:47,210 And now it's going to be very challenging. 129 00:07:47,250 --> 00:07:48,570 It's not going to be challenging at all. 130 00:07:48,610 --> 00:07:51,170 We're going to say memory equals true. 131 00:07:51,890 --> 00:07:54,810 And we are going to then just do exactly that. 132 00:07:54,850 --> 00:07:57,490 So crew was a little bit off base with the memory equals part. 133 00:07:57,490 --> 00:07:58,690 But it's got the rest of it right. 134 00:07:58,690 --> 00:08:00,330 That's all you need to do. 135 00:08:00,490 --> 00:08:04,970 You set the long term memory the short term memory and the entity memory. 136 00:08:04,970 --> 00:08:07,010 And we are almost done with memory. 137 00:08:07,010 --> 00:08:08,610 Just just as simple as that. 138 00:08:09,170 --> 00:08:14,510 And I said almost because there is just one or rather two very small extra changes we need to make. 139 00:08:14,550 --> 00:08:21,910 We need to go back up in the module to where we created these agents the trading company, the financial 140 00:08:21,910 --> 00:08:23,430 researcher and the stock picker. 141 00:08:23,550 --> 00:08:25,310 And we need to give them memory. 142 00:08:25,350 --> 00:08:29,830 Now what we want is for the trading company finder to have memory. 143 00:08:29,830 --> 00:08:32,190 And we just do it by saying memory equals true. 144 00:08:32,510 --> 00:08:36,390 We don't actually want the researcher to have memory because we want it to go and do research every 145 00:08:36,430 --> 00:08:36,950 time. 146 00:08:37,110 --> 00:08:41,830 But we do want the stock picker to have memory, because we don't want it to recommend the same thing 147 00:08:41,870 --> 00:08:42,550 more than once. 148 00:08:42,550 --> 00:08:48,470 And I don't know if you remember, but in the prompts, in the YAML files, I said a couple of times, 149 00:08:48,790 --> 00:08:52,590 uh, don't, don't recommend the same stock twice and things like that. 150 00:08:52,590 --> 00:08:56,190 And, and surface new companies for the for the trading company finder. 151 00:08:56,190 --> 00:09:01,070 And that would normally be the final change you need to make is go back and make sure that your instructions 152 00:09:01,070 --> 00:09:05,750 and your YAML files are very clearly making sure that it will take advantage of memory. 153 00:09:05,750 --> 00:09:11,580 Because remember, whilst memory, these abstractions are trying to make memory seem quite magical and 154 00:09:11,580 --> 00:09:12,780 taken care of for you. 155 00:09:12,780 --> 00:09:19,660 At the end of the day, memory just means more stuff shoved into the prompt, more relevant context 156 00:09:19,700 --> 00:09:24,220 put into the prompt so that when you call an LLM it has knowledge. 157 00:09:24,260 --> 00:09:30,220 It's in the input is included information about prior conversations or about prior information that 158 00:09:30,220 --> 00:09:31,020 it retrieved. 159 00:09:31,340 --> 00:09:34,660 So with that we have set up set up the memory. 160 00:09:34,820 --> 00:09:39,860 And we are now going to bring up our terminal and then run this. 161 00:09:40,220 --> 00:09:43,900 And so as usual I go into already in the stock picker. 162 00:09:43,900 --> 00:09:47,540 So all I have to do is type a query I run and we'll be good to go. 163 00:09:47,740 --> 00:09:49,060 Let's see what happens. 164 00:09:49,260 --> 00:09:55,420 So just right off the bat, we expect it to be able to take advantage of memory without needing anything 165 00:09:55,420 --> 00:09:55,820 more. 166 00:09:55,980 --> 00:09:59,540 What we should see is it should create a memory directory in here. 167 00:09:59,540 --> 00:10:02,260 And it just has there is a memory directory. 168 00:10:02,460 --> 00:10:04,020 Stuff is going on. 169 00:10:04,060 --> 00:10:10,340 My computer is hard at work and I can see already within memory there is a Croma database that's being 170 00:10:10,340 --> 00:10:10,980 created. 171 00:10:10,980 --> 00:10:12,960 There is a long term memory. 172 00:10:12,960 --> 00:10:18,480 If I expand this, there is indeed a database that's been a SQLite database that's been created there. 173 00:10:18,480 --> 00:10:26,440 And so things are happening and we can see that, that, uh, various companies are being surfaced by 174 00:10:26,440 --> 00:10:33,320 the market watcher and more is going on and we will, uh, let this thing run. 175 00:10:33,360 --> 00:10:37,640 My computer's hard at work, and I will see you when we have a conclusion. 176 00:10:38,080 --> 00:10:39,640 And that definitely took a bit longer. 177 00:10:39,640 --> 00:10:43,040 It was going around around the houses a little bit, but it completed it. 178 00:10:43,040 --> 00:10:44,760 Recommended Microsoft this time. 179 00:10:44,880 --> 00:10:46,680 Remember, don't use this for real decisions. 180 00:10:46,680 --> 00:10:51,080 But it was, uh, entertaining to see it at work and bouncing around between the different agents. 181 00:10:51,200 --> 00:10:56,280 And whilst we don't have as much visibility into what's happening in terms of its use of memory and 182 00:10:56,280 --> 00:11:02,320 what context got provided, we can see that it's certainly built and populated different data stores 183 00:11:02,360 --> 00:11:09,320 and both the short term memory, the long term memory and the entity memory in the Rag chroma data store 184 00:11:09,320 --> 00:11:10,480 that's been created there. 185 00:11:10,520 --> 00:11:15,550 And the main point I want to get across is that, of course you can see the benefits of what this brings 186 00:11:15,550 --> 00:11:15,750 us. 187 00:11:15,750 --> 00:11:23,590 It was so easy to set up quite a complex situation, multiple types of memory with both vector similarity 188 00:11:23,590 --> 00:11:25,710 queries and SQL queries too. 189 00:11:25,750 --> 00:11:27,430 And we didn't need to know anything about it. 190 00:11:27,470 --> 00:11:29,830 We simply created the objects. 191 00:11:29,830 --> 00:11:33,950 The short term long term entity memory objects passed them in. 192 00:11:34,070 --> 00:11:40,190 And then we told our agents we turned memory on by saying memory equals true for the agents that we 193 00:11:40,190 --> 00:11:41,910 wanted to remember things. 194 00:11:42,350 --> 00:11:45,510 And that is a wrap on the stock picker project. 195 00:11:45,550 --> 00:11:49,350 We saw a lot of different aspects of Q AI with this project. 196 00:11:49,390 --> 00:11:52,150 We reminder, we saw structured outputs. 197 00:11:52,310 --> 00:11:55,870 We saw our own homegrown tool as well as Serpa. 198 00:11:56,190 --> 00:12:01,750 We also used the not the sequential but the hierarchical process. 199 00:12:01,950 --> 00:12:09,750 And now we have added in the memory feature all, all the three main types of memory in there as well. 200 00:12:09,790 --> 00:12:14,390 And that is a nice tour of a lot of the functionality in crew.