Introducing Strands Agents, an Open Source AI Agents SDK — Suman Debnath, AWS
00:00:00.000 | -
00:00:03.000 | All right, so what we are gonna do is,
00:00:16.280 | I'll quickly go through a couple of demos, that's all.
00:00:20.000 | That's the plan.
00:00:21.560 | So Strandset is an open source SDK,
00:00:25.560 | which has only one motivation,
00:00:29.420 | is to make your agents as simple as possible,
00:00:33.420 | without any scaffolding.
00:00:34.900 | The only thing that is needed is tool and the model.
00:00:39.900 | That's all, there is no scaffolding,
00:00:43.080 | you don't have to dictate what the agent should do,
00:00:47.220 | and the idea is, since the models are becoming
00:00:51.460 | more and more intelligent, we want the model to take
00:00:55.460 | the reasoning part of our agents.
00:00:58.340 | We don't want to scaffold with a lot of prompts,
00:01:01.000 | a lot of system prompts, a lot of backgrounds,
00:01:03.820 | and all that.
00:01:04.440 | So that's all to it.
00:01:07.920 | And it is also integrated with various
00:01:10.560 | third-party providers, like LangFuse, LightLLM.
00:01:17.720 | You can use any model of your choice.
00:01:19.820 | You don't need to use Bedrock hosted models, but you can go ahead with any of the model of your choice using LightLLM.
00:01:27.360 | And also you can use OLAMA if you want to test it out locally.
00:01:32.720 | Now, this is the agentic loop that we have in strands.
00:01:38.200 | As I mentioned, it's just the models and the tools, that's all.
00:01:41.980 | And that is actually the two strands in this logo of strand denotes.
00:01:47.380 | One is for model, one is for tool.
00:01:49.200 | I just happened to know this yesterday.
00:01:51.360 | So, yeah.
00:01:52.360 | Okay, so now let's see a demo, how you can create a strand agent.
00:02:00.240 | So the demo that we are going to have now, it's very simple.
00:02:03.700 | We are going to read one file from our disk, and then we will generate a summary of that.
00:02:09.740 | Then we will write that summary into our local drive.
00:02:14.340 | And then we will also tell the agent to speak out the result.
00:02:18.920 | So basically, we are trying to add the voice aspect of this, okay?
00:02:22.860 | So now there are three things we are doing: reading a file, writing a file, and generating a speech.
00:02:30.120 | All of these three, we are doing it using the default tools from strands.
00:02:34.700 | So you don't have to write any custom tool for this.
00:02:36.980 | You just have to install a strands tool, and boom, you can get started with this, okay?
00:02:44.560 | So just remember this use case.
00:02:46.620 | I have recorded this because I heard that the Wi-Fi is pretty fantastic here.
00:02:50.940 | All right, so let's see how it goes.
00:03:00.860 | So here, the first thing we need to do is we need to install strands.
00:03:06.620 | So that is what we are doing here.
00:03:09.580 | We are just installing the strands agent, and then we are going to install strand tools.
00:03:16.120 | So this is a simple pip install, nothing fancy here.
00:03:19.540 | And when you install the strands tool, it gives you a lot of default tools.
00:03:23.720 | You can always make your own custom tools, but this is how you can create or use out-of-the-box
00:03:30.200 | tools.
00:03:31.200 | These are the three tools that we are going to use.
00:03:34.160 | So now we are going to write the application.
00:03:44.120 | So the first thing that we need to do is we need to import strands, and by default, it uses
00:03:50.000 | bedrock model, which is Cloud 3.7, under the hood.
00:03:54.120 | But if you don't want to use that, you can always use any model of your choice.
00:03:57.080 | But by default, it uses bedrock.
00:03:59.080 | I am trying to show you how you can define the model, but later on when we create agents,
00:04:05.540 | you will see that you don't have to give model ID.
00:04:09.120 | By default, only it should work.
00:04:10.880 | But I just want to show you, if you want to define your own model, how you can define the
00:04:15.040 | model.
00:04:16.040 | In case you want olama, you can use olama there.
00:04:20.000 | And now this is a system prompt.
00:04:21.500 | When I share the GitHub repo, try to do this, delete the system prompt, it should still work.
00:04:26.620 | But I just wanted to give you an intuition of how things work under the hood.
00:04:31.360 | And if you see here, all we are doing here, we are giving the model ID, system prompt, and
00:04:35.280 | the tools.
00:04:36.280 | And now we are asking the same question, read this chapter number 10, summarize that,
00:04:43.240 | and write it in a markdown file, and speak out.
00:04:46.900 | Now to listen to the speech, let me increase the volume.
00:04:53.000 | So now it is reading the file, then it is trying to write the file after creating the summary,
00:04:58.040 | and now it should speak out.
00:04:59.520 | The human eye functions like a camera, with light entering through the cornea.
00:05:03.600 | So you get the idea, right, so how these things work.
00:05:07.780 | Now there is one more demo I have, which is around MCP with strands.
00:05:14.880 | Because without MCP and agents, I was told that you cannot have any talk in this conference.
00:05:20.140 | So.
00:05:21.140 | All right.
00:05:22.140 | How many of you have heard of 3Blue1Brown YouTube channel?
00:05:28.140 | Oh, wow.
00:05:29.140 | So you know what we are talking about.
00:05:31.300 | So we are trying to build animations and videos like what we see in 3Blue1Brown.
00:05:37.480 | It is powered by one of the library called Manim.
00:05:40.480 | So I have been working with this Manim for a couple of years to, if you are from science
00:05:47.740 | and mathematics background, eigenvalue, you know, eigenvectors, SVD, these are the concepts
00:05:53.480 | which are very easily we can explain using visualization.
00:05:56.900 | So what we are going to do now is we are going to create a strands agent using an MCP server,
00:06:06.080 | which will help me to generate these videos using Manim.
00:06:09.920 | Manim.
00:06:10.920 | Manim.
00:06:11.920 | With me so far?
00:06:12.920 | Yeah.
00:06:13.920 | So this is how we can do.
00:06:17.100 | Assume that we already have an MCP server which can generate this video.
00:06:22.100 | Okay.
00:06:23.100 | Just assume that.
00:06:24.100 | It is not here, but it is already there.
00:06:27.100 | Okay.
00:06:28.100 | So now what we are going to do is using strands, we are just going to call the MCP client and
00:06:37.180 | make use of that server.
00:06:38.280 | So in this case, I am importing the strands agent and the MCP client.
00:06:45.080 | The next thing that we are going to do is we are going to create the MCP client.
00:06:49.120 | If you see here, this is my MCP server path where I have returned the code to execute that
00:06:58.300 | MC, Manim code.
00:07:01.180 | So that, this particular code I am not showing, but after this I will show it to you what that
00:07:05.800 | code is doing.
00:07:07.200 | But that's all.
00:07:08.880 | I don't have to do anything else.
00:07:11.300 | Now, when I create an agent, in the previous example, if you remember, we mentioned that
00:07:16.880 | agent then tools equal to read file, write file, and speak.
00:07:23.180 | In this case, I am not using any of the default tools.
00:07:26.020 | My tool is nothing but the MCP server.
00:07:30.460 | And for that, I am just using tools equal to Manim MCP client list tool sync.
00:07:36.500 | So this will broadcast all the tools that the MCP server has to my client.
00:07:43.340 | So that is nothing but the strands agent.
00:07:46.220 | With me so far?
00:07:47.220 | Yes?
00:07:48.220 | Perfect.
00:07:49.220 | And at the end, we are just going to give the prompt.
00:07:51.780 | So in this case, you can give any prompt of your choice.
00:07:54.660 | But here we are just saying that create a visualization for this cubic equation within the range of x minus
00:08:01.260 | 3 to x equal to 3.
00:08:04.440 | If you want to try this with something else, you can always give, like, explain me byte-pair encoding in a visual way.
00:08:11.580 | Or you can say, explain SVD, explain how two metrics of 2 x 2 perform a metric multiplication, or whatever, right?
00:08:24.460 | So now let's see this.
00:08:25.820 | So for this, before I start the MCP client or the strand agent, I have to make sure that server is up and running, right?
00:08:33.440 | So in here, I'm just making the MCP server up so that this is available in that local host.
00:08:40.580 | And now I'm running this app.py, which contains the code which we have seen a while back.
00:08:47.720 | So finally, we can just give the prompt.
00:08:49.720 | So this is the same prompt that we have seen in the previous slide.
00:08:52.720 | And let's see what it generates.
00:08:54.860 | And if you notice, we have not mentioned any scaffolding, kind of, like, system prompt and all that.
00:09:05.860 | We want the model to reason about this and figure it out of its own.
00:09:13.860 | So I have not done anything.
00:09:16.000 | And it is not Photoshopped.
00:09:18.000 | It just came up like that.
00:09:21.000 | And you can actually define the duration of the video.
00:09:28.140 | Let's say 30 seconds, 10 seconds, and all that.
00:09:32.140 | And if you have worked in Manim to generate this code, it takes some time, right?
00:09:37.140 | So it's not that straightforward.
00:09:40.280 | Okay?
00:09:41.280 | So that's about the integration with MCP and Strand.
00:09:46.280 | And before I show you a few of the resources, let me show you the code.
00:09:52.280 | No one pings me while I'm working, but everybody pings when -- okay.
00:10:04.280 | All right.
00:10:05.280 | So this is the -- so this is the MCP server that I have.
00:10:14.280 | It just -- I have imported fast MCP and this is the -- this is a couple of the tools that
00:10:27.380 | I have defined.
00:10:28.620 | And if you see this here, the way that you can create an MCP tool is just wrap it around
00:10:33.040 | with an MCP.tool decorator, right?
00:10:35.660 | So this becomes -- these two are the tools for my MCP server.
00:10:40.960 | And when I come here -- when I come here, and if you look at the main function -- so this
00:10:50.080 | is the function where I'm going to call that using strands agent and MCP client, right?
00:10:56.740 | So I -- what I've shown in the presentation is just a few lines from this main function.
00:11:02.740 | But we are doing a little bit more before just calling the agent.
00:11:08.440 | We are listing the tools, but before that, we are also giving some examples.
00:11:14.900 | And if I show you this example, I'm just giving one example of how to create, you know, and
00:11:26.320 | construct in using Manning.
00:11:28.420 | That's all.
00:11:29.420 | You can think of it like one-shot prompting.
00:11:31.560 | It's like that.
00:11:33.940 | But my -- as far as the agent is concerned, it's just one line of code.
00:11:39.680 | You know, we just list the tools.
00:11:43.020 | And then here, we are just calling the agents.
00:11:45.380 | That's all.
00:11:46.380 | And then we have a while true loop so that we get an interactive chat kind of an experience.
00:11:53.020 | And in our first demo -- not first demo -- I want to show you one quick thing is how you
00:12:04.980 | can create a custom tool.
00:12:07.480 | Okay?
00:12:08.480 | This is -- I'm just jumping around a couple of things because we just have two, three minutes.
00:12:14.480 | It's no less than a time bomb, trust me, when I see this.
00:12:18.240 | So let's say you have any function -- let me just quickly scroll down and show it to you.
00:12:29.480 | Yeah.
00:12:30.480 | So if you have any function, and you want to create that function, you know, turn that function
00:12:40.380 | into a tool, all you have to do is this -- let's say this is a function.
00:12:50.360 | This is a different example.
00:12:52.260 | It's there in the GitHub, but I will not explain that.
00:12:54.160 | But the idea is, let's say you have a function.
00:12:56.060 | You want to convert that into a tool.
00:12:57.960 | All you have to do is just import tools from here and decorate it with tools.
00:13:03.900 | Once that is done, this retrieve from Quadrant will become a tool for you.
00:13:09.960 | So next time when you create an agent, when you create an agent, you just give that as a tool.
00:13:17.040 | And now here, in this list of tools, you can give your custom tool as well as your default
00:13:21.860 | tools, whatever we have -- read file, write file, speak, and so on and so forth.
00:13:26.440 | Okay.
00:13:27.440 | So that's about it.
00:13:29.440 | Now, if you want to learn more about strands, we have a GitHub repository.
00:13:35.380 | So this is the launch blog.
00:13:38.440 | And this is the documentation, which is nothing but strandsagent.com.
00:13:44.380 | And this is the GitHub link.
00:13:47.380 | Okay.
00:13:48.380 | And if you have any questions, feel free to, you know, ask me.
00:13:52.440 | I'll be around.
00:13:53.440 | And we have a booth inside as well.
00:13:55.380 | There are plenty of demos with strands, with Lambda, MCP and, you know, to get started.
00:14:03.320 | This is an open source project.
00:14:04.320 | Feel free to raise a PR if you have something.
00:14:07.380 | If you have built something into the samples code repository within strands, GitHub.
00:14:12.380 | And, yeah, share what you built.
00:14:15.380 | And if you have any feedback, let us know.
00:14:17.380 | Thank you so much.
00:14:18.380 | .
00:14:19.380 | .
00:14:24.380 | .
00:14:24.380 | .
00:14:24.380 | .
00:14:24.380 | .
00:14:24.380 | .
00:14:24.380 | .
00:14:24.380 | We'll be right back.