VectorShift Just Launched AI Agents - Here's Everything New

VectorShiftRecent📅 2025-06-18

Watch on YouTube ↗

AI Agent Architecture

┌─────────────────────────────────────────────────────────────────────┐
│                       AI AGENT FLOW                                 │
├─────────────────────────────────────────────────────────────────────┤
│                                                                     │
│   ┌─────────────┐                                                   │
│   │    USER     │                                                   │
│   │   QUERY     │                                                   │
│   └──────┬──────┘                                                   │
│          │                                                          │
│          ▼                                                          │
│   ┌──────────────────────────────────────────────────┐             │
│   │                   AI AGENT                        │             │
│   │  ┌────────────────────────────────────────────┐  │             │
│   │  │              LLM BRAIN                      │  │             │
│   │  │   1. Understand query                       │  │             │
│   │  │   2. Plan approach                          │  │             │
│   │  │   3. Select tools                           │  │             │
│   │  └────────────────────────────────────────────┘  │             │
│   │                      │                           │             │
│   │         ┌────────────┼────────────┐              │             │
│   │         ▼            ▼            ▼              │             │
│   │   ┌──────────┐ ┌──────────┐ ┌──────────┐        │             │
│   │   │  Tool 1  │ │  Tool 2  │ │  Tool 3  │        │             │
│   │   │ Web API  │ │  Files   │ │  Code    │        │             │
│   │   └──────────┘ └──────────┘ └──────────┘        │             │
│   └──────────────────────────────────────────────────┘             │
│          │                                                          │
│          ▼                                                          │
│   ┌─────────────┐                                                   │
│   │  RESPONSE   │                                                   │
│   └─────────────┘                                                   │
│                                                                     │
└─────────────────────────────────────────────────────────────────────┘

Agent Components

• Agent configuration
• Tool integration
• Memory management

Transcript

[00:00] Vector Shift has made it incredibly easy to build AI agents. This single agent has access to pretty much all the functionality you'll ever need. It's able to go online and perform deep research. It can answer questions from a custom knowledge base, integrate with third-party applications, and call other vector shift pipelines as well. Vector shift has always been a fantastic tool for building deterministic workflows using LLM. But one feature that was always lacking was the ability to add AI

[00:31] agents into these workflows. And I'm referring to agents that can solve reason call tools and work with other agents to solve a problem. So in this video, you'll learn how to build your very first AI agent in Vector Shift. From the Vector Shift dashboard, you will notice this new menu item called agents. This is where you'll create your agents without writing any code or dragging nodes onto a canvas. Let's start by creating a new agent and let's give it a name like Jarvis. Let's create

[01:01] this agent. And now we can simply set up our agent any way we want. On the left hand side we can change between basic config lm config where we can select our provider and the model that we'd like to use. And of course we can assign tools to this agent as well. Let's start with the basic config. Here we can set up the different inputs to the agent. By default, we can send text to the agent. But if we wanted to, we could also add another input. And maybe we wanted to send an image or a file to the agent as

[01:33] well. For this demo, we'll simply go with text input. Scrolling down, we can set up the agent instructions. And this is effectively the system prompt, which will change the behavior and the role of the agent. So we could add something like your name is Jarvis. Then on the right hand side we can test this agent. So let's send hey let's run this and we can see the output popping up over here. Cool. Now let's start to assign tools to this agent. Within tools we can assign things like knowledge bases and

[02:05] pipelines or we can also integrate with APIs scrape websites perform a web search using Google search or XAI search and of course we can integrate the agent with external systems under integrations. Let's start by adding a custom knowledge base. So let's click on query knowledge base and I'm going to assign a knowledge base that contains information about my restaurant called the oak barrel. So under tool description we can tell the agent when to use this tool. So let's say use this

[02:36] tool to retrieve information about the oak and barrel restaurant. Then under knowledge base let's create a new knowledge base. Let's call this oak and barrel QA. Let's create this. And when we close this popup, this will take us to the knowledgebased page where we can upload documents, integrate with external systems or scrape data from a website. I'm simply going to upload a file. So I'll select this word document

[03:07] containing Q&As about the restaurant. Then let's click on finish. And in the status, I can see this was successful. So let's go back to Jarvis and that is it. Just ensure that you select the knowledge base from this dropdown and everything else should be sorted. We can also rename this knowledge base. So let's say oak and barrel KB for knowledge base. Right? We can see that tool was added to the list of tools and we should be able to test it already. Let's ask what are the current specials

[03:41] at the oak and barrel. Let's send this. And indeed, we get this answer back from the knowledge base. Perfect. Let's add another tool. This time I want to give the agent the ability to go online and retrieve realtime information. So to do this, let's go to popular. And now we can either add Google search or XAI search. But I personally find XAI returns better results. Let's add this tool. I'm going to rename this to web

[04:11] search. Let's save this. And we don't have to change anything else. We can already test this. So let's ask what is the current weather in New York. Let's run this. This is the kind of information that wouldn't be available within the model's training data. So the model will realize that it needs to perform a web search. And there we go. We get the current weather back using real-time data. I would also like to give my agent the ability to scrape data from a website. So let's do that by

[04:44] going to add. Then under popular, let's select URL loader. I'm going to rename this to scrape URL only. We can leave the tool description and everything else on its default value. And we should be able to test this already. So, let's grab this article from OpenAI. I'll just copy the URL and let's ask our agent, please extract the content from this page and summarize it. Let's paste in

[05:14] that URL and let's run this. And the agent should now use the scrape URL tool to extract information from this page and then summarize it for us. And that was actually really quick. So, it's extracted all of this information from that page and summarized it. Now I do want to take this one step further. I want the agent to extract this data, summarize it and then email the report to me. So to add email integration, let's click on add. And under integrations, let's look for Gmail. Then

[05:45] for the description, let's say use this tool to send emails. Then under integration account, you can simply click on this button and then click on add integration and authenticate your Google account. I've already done that. So, I'm just going to select this. And under action, let's select send email. Of course, you can just create a draft email or a draft reply, whatever you want, but I'll be brave and just send the email. For the format, we can select

[06:15] between text and HTML. I'll just leave it on text. And then for the two address, the subject line, body, and attachments. I'll let the AI agent decide what to do. And believe it or not, that's all we have to do to add email integration. So I'm going to extend this prompt to say email the summary to leonfunsdev@gmail.com. Right, let's run this. All right, so we get our summary. And if we go to our email, I can see that I've received an

[06:47] email and there you go. Our agent was able to send an email containing the summary. Our agents are also able to call other pipelines within vector shift. So let's add another tool and this time let's select pipeline and let's say use this tool to create blog posts. I'm going to rename this to blog post generator and let's save this. And under pipeline let's create a new pipeline. Let's call this blog post

[07:20] generator. Let's create this. And for this, let's add an input node, which I'll rename to topic. Then under AI, let's add the open AI node. For the system prompt, let's say your role is to create an outline for a blog post based on the following topic. Cool. Then for the prompt, we can just enter topic. Then we can grab a variable

[07:50] by typing double curly braces. We'll select topic.ext. All right. So this will then generate the outline. Let's add another llm node. For this one system instruction, we'll say create a blog post based on the following outline and topic. Cool. Then for the prompt, let's say topic, which is the topic node.ext text and the outline is the OpenAI node response. And

[08:21] finally, let's add our output node and the output will be OpenAI1 response. In order for our agent to access this pipeline, we do have to publish it. So, let's click on deploy changes and let's click on deploy. Cool. Let's go back to Jarvis. Just make sure that you've got that pipeline selected. And that's actually all we have to do. So let's try it out. Let's say create a blog post for the following topic. The impact of AI agents in the workplace.

[08:55] Let's run this. And our agent should now use the pipeline to generate this response. We get this blog post back with all the different sections like an introduction, what are AI agents, and much much more. Cool. So, if we wanted to, we could have asked the agent to email this article to us, but I think this demonstrates the point. Another popular use case for agents is the ability to perform deep research. We can do that in vector shift as well, and it's super simple. Let's add a new tool.

[09:25] Let's select pipeline. I'm going to rename this tool to deep research. Under tool description, let's say use this tool to perform deep research and generate detailed reports. Under pipeline, let's create a new pipeline. Let's call this deep research. Let's click on create. And this is actually a very simple pipeline. Let's add the input node. I'll just rename this to topic. Then under AI, we're actually going to use a deep research model from

[09:58] Perplexity. So let's add the perplexity node. For the model, let's select sonar deep research. And for the system prompt, let's say generate a detailed research report based on the following topic. Then in the prompt, let's say topic is equal to topic.ext. Cool. I'm just going to rename this note to research. And finally, we just have to return the output by using the output node. So in output let's say research

[10:32] dot response and again in order for the agent to access this pipeline we do have to deploy it. So let's click on deploy. Let's go back to our agent. And that is actually all we have to do. This research model can take quite a while to execute. So it's a good idea to add this to another pipeline which could run in the background. So what I'm going to do is deploy this agent so that other pipelines can access it. Then let's go back to pipelines. Let's create a new pipeline. Let's start from scratch. And

[11:04] let's call this pipeline my agent. Then let's add our input node. And I'm going to call this user message. Then under objects, let's add the agent node. And in the agent dropdown, let's select Jarvis. And for the input, we'll simply grab it from user message.ext. And finally, we can return the agents output by adding the output node. And for output, let's select agent.output. And believe it or not, that's all we

[11:34] have to do. Let's click on run. And let's say create a research report on the impact of AI agents in the workplace. Email the report to leonfaledev@gmail.com and let's run this. The perplexity research model can take a few minutes to complete. So I'll see you soon. And while this is running, I do want to show you a small tip to see whether or not

[12:04] your pipelines are being called by the agent. If we go back to the vector shift dashboard under pipelines, you can click on run history. And this is where you can see all the pipelines that's currently being executed. So of course we used this pipeline to trigger the agent and the agent in turn is executing our deep research pipeline by itself. And we can see that research pipeline is still running. So that's a useful tip to see what's going on under the hood. Another tip for seeing what this agent

[12:35] is doing under the hood is by opening up this trace view. And here we can see the agent node. If we expand this, we can see the agent that's being run. And if we open this up even further, we can see the models that's being used. We can also see that there's a pipeline being executed. And if we expand this, we can see it's a deep research pipeline that was run along with its inputs and outputs. We can also see that the Gmail tool was run as well since we asked it

[13:06] to send us an email. And cool, we get this response saying the report was created and sent to our email address. I don't know about you, but I think agents is a very welcome addition to Vector Shift. Let me know in the comments if you'll be adding agents to your projects in Vector Shift. And also check out my other Vector Shift videos over here. I'll see you in the next one. Bye-bye.