Name: Mistral AI Agents
Uploaded: 2024-12-18T12:42:09.853Z
Duration: 2 min 33 s
Description: A feature by Mistral AI enabling the creation of custom AI agents for automating workflows and tasks.

mistral-ai-agents

A feature by Mistral AI enabling the creation of custom AI agents for automating workflows and tasks.

Mistral AI Agents

This is amazing. Now we are going to see about Mistral AI Agents. AI agents are going to be the future. If you see here, we've already accomplished level one and level two chat bots and reasoners. Now level three is agents, so it's vital to focus on AI agents and do not miss this opportunity of creating applications with AI agents. 

AI agents are nothing but independent AI systems working together to solve complex tasks. As an example, here we have three different agents. The first one is the planning agent. So when you provide a task to the planning agent, it is able to create a list of steps to complete the task. Then those list of steps is forwarded to the Python agent, who is going to write Python code to execute each individual step. Then that output is sent to the summarization agent, who is going to summarize the whole results. 

Recently, Mistral AI released AI agents, which means on their platform we are able to create an agent and assign up specific tasks. For example, if it's a French agent, you can tell it you are a French-speaking virtual agent designed to answer your questions in French only, no matter the language of the question. And also we provide some examples such as.

This and then it is able to respond only in French even though the question was in English. So this is one agent that that is French agent. Similarly, we are going to create three different agents and make them work together. 

So by the end of this video, you will learn how you can create agents, how to assign task, how to run agents, how to integrate that with your own application. That's exactly what we're going to see today. Let's get [Music] started. 

Hi everyone, I'm really excited to show you about Mistral AI agents. In this, we are going to create multiple agents that is multi-agent approach where we'll be creating separate agent like this, then assign task to them and make them work together. 

But before that, I regularly create videos and request to Artificial Intelligence on my YouTube channel, so share AI Agent Store with colleagues. First, we are going to go into the platform that is Mr ai's own platform. There you got agents, so click on create agent and there you can choose the model of your choice. Then first, we are going to create three different agents.

So first step creating agents first is the planning agent. So here just mention the name planning agent. Then you need to provide the system prompt. So the system prompt I'm giving is you are a data analytical planning assistant given a data set and its description. Your task is to provide specific and simple analysis plans, detailed instructions, and suggested python code that can later be given to separate python agent. 

So after giving this prompt, click on deploy. You can access this agents via API and also via the chat interface. But in my case, I'm going to keep only API. Second agent is the python agent, same as before providing the name for the agent python agents. Setting the model name Mr L 2. Here I'm saying you are a python coding assistant that only outputs python code without any explanations or comments. 

And I'm providing some example here. If the user put is how can I sort a list of numbers in ascending order, then the answer should be like this only with the python code. Then after providing this example, click deploy. Finally, the summarization agent, same as before providing the name for the summarization agent, providing the model name and the.

Instruction you are analysis summarization assistant given a data set description and analysis results provide an analysis report that's it. And click deploy now we have the summarization agent available via API now. Step number one completed. 

Now we need to write the code so step number two is create the code and assign tasks to those agents. Pip install mral Ai and pandas and then click enter. Pandas is used for data analysis. Next export your mystal AI API key like this and then click enter. After this let's create a file called app.py and let's open it. 

Inside the file first we are going to import OS then Mistral AI and import IO. Next we are getting the mral API key. Let's create the client mral with API key. Next we are going to provide three different agents planning agent ID summarization agent ID and the python agent ID. You can get all these agent IDs from the portal here by clicking agents and you'll have all the agents created here and you can see the IDS listed just on the right hand side. 

So you can just copy the IDS by clicking this icon. I've provided those is here. Now we need to set up these three agents and assign tasks to those agents first.

Planning agent so in the file let's create a function called run analysis planning agents. So this is just calling the client agents.

Compete function there we are providing the agent ID and the message that's the task which we are going to provide here. So this is the main function which we are calling under this function. Next we are going to provide the query with the data and all other information about the data set. So the main task is to providing the data set like this and giving you a brief description of the data set and asking those agents to analyze the data within the data set. So that is a key task. 

So now these three agents are going to work together. So now I've given the data set. So the data set looks like this state number of drivers percentage of drivers in collision and other information such as car insurance premium Etc. Next calling the planning agent function and printing the results that's it first agent completed. 

Next let's create the python agent which means Define the python agent and assigning task here. I'm going to create a class called python agent workflow. In this python agent workflow class first it goes through the workflow. So in the workflow it consists of multiple different steps to complete. Then each step is processed in this process step request and.

After that it goes to the checking code and extract code. So the extracted code is then run to get the results. So in simple terms this gets the python code and run it and return the response. That's it. 

So you can see the full code here. I will put all the code in the description below. Next we need to capture the different conversation or the different response which will happen within the python agent. So I'm going to add a few more code with the T clause and this will automatically get all the conversation or the API calls which happened inside the python agent workflow because it contains multiple steps. 

So everything is captured using this and the final output is saved in the captured output. That's it. Now final agent that is a summarization agent same as before response equals client agents.

Complete and providing the summarization agent ID with the initial question all for the query and the captured output from the python response agent. So we sent all those information to the summarization agent and finally we're going to print the response result equals response choices message content and printing the result. 

That's it. So as a quick summary first we provided this data with a description about the data to analyze to the first planning agent. The planning agent will create the list of steps to follow and send it to the python agent. The python agent will create code extract the code run the code and send the response back to the summarization agent. 

Finally, the summarization agent will summarize the whole response of the conversation. Now I'm going to run this code in your terminal Python app.py and then click enter. First is going to the planning agent and the planning agent is creating steps planning the steps totally three steps. 

Step number one load the data set from the given URL and it's giving the suggested code. Step number two inspect the first few rows of the data set to understand its structure using DF do head. Step number.

Three check for missing values in the data set using isol and sending all this information to the python agent. The python agent step number one it's using the code extracting the code running the code and you can see code executed successfully. Then it goes to step number two inspect the first few rows of the data set. Then checking for missing values in the data set. Then finally it gives us some summary of analysis about the data set provided. 

Similarly, you are able to provide any data set to these list of agents and they are able to analyze for you which is going to make your work more simple. That is the power of AI agents. I'm really excited about this. I'm going to create more videos similar to this so stay tuned. I hope you like this video. Do like, share AI Agent Store with colleagues, and thanks for watching.

Note, that there may be slight differences in text and video. Watch video for the original content.

Mistral AI Agents Video