vanna-ai

An open-source AI-powered SQL agent enabling users to interact with databases using natural language queries.

Vanna AI

Today we're going to explore Vanna AI. Vanna AI is a powerful tool to write SQL for you using national language prompts. Using this amazing tool, you can get insights to your database and you can actually train your own rag model. 

So if you're new to the concept, you can always go to their documentation. Here on their documentation, you can find their quick start sample data which you can run on Google Colab or you can open it in GitHub. Then you can actually interact with it. 

And that's not all. You can even go on and go with different databases like Postgres, DuckDB, Microsoft SQL, MySQL, and much more. So we already know that Vanna is quite flexible and it provides different options. 

But how to get started with W? All you have to do is sign up with W and you'll be directed to their dashboard which would look something like this. Now W gives you a free plan in which you get the charge 3.5 LLM using which you can train your rag model over at Utata. 

Now this is more than enough if you have a beginner level project and I totally recommend that you go with the free plan for starters. Later on, if you think that this is worth it, you can go with the premium version or you can go with the Enterprise version as well. 

Now a few things that you need to know is in the usage tab you get the summary about your usage which if you've been using the rag model, the AI model, and the number of requests as well as you get your own API key and you can create your own rag models. 

So I've already created my model Skill CB here but of course it is not trained yet. You can add training data to it and create your own rag model again. 

Now that was all the basic knowledge that you need to know about W AI. But now let's get started with the startup code here. So what I'm going to do is I'm going to actually copy this code snippet here. This is a sample code snippet which will run the Chino Ragu as well as the Chino SQL database. 

Then you can ask it questions using natural language prompts. So I'm going to go to my Google Colab and here I can run that. But one more thing that I want to tell you is that you can actually run the same thing over at Jupyter Lab as well. 

So it is really up to you what you prefer. For me, I'm going to go with both of them and I'm going to show you how you can do it in both of these amazing platforms. Now as we can see that our code SH is already. 

Please share the AI Agent Store with friends.

Executed and we installed W. Then we went on and connected to the rag model and here are the result we get the SQL cery as well as we get the D table form. There is even a graphical representation using which you can actually visualize your data. So that was is using Vanna inside cab but you can actually launch the Vanna of FL cap. 

If you go all the way up you can see in the sample code snippet you actually go on and run the wof flas cap. So all the way down you see a link here. So let's go and open up this link and here is your W with a GUI. Now you can ask different questions to your rag model which is specifically trained on your database or you can add trading data as well. 

All you have to do is add trading data and here you will have to select the type of trading data that you're adding. Either it is going to be SQL queres or documentation or it is going to be the ddl statements or cues. Once you're done adding it, it will pop up right here where you can actually view all the train data that you have already added. The questions as well as the Curious that is really simple and easy to do to drain your rag motor using GUI and ask question questions. 

So let's go and ask a question using this GUI. So I'm going to go and select this question here which five cities have the most customer. So it's actually going to go and give me the query. So it went on and gave me the cery of how it accepted this certain data out of the database. Then it gives me the data in a tabular format as well as the data in a CSV file which I can download. 

There is a graphical representation with which we can actually visualize our data. Now that's not all you can even redraw this chart and you can even get the summary of all of this which is just mind-blowing how much information we get using just one simple national language promt. So that was the GUI version of Vanna but now I'm going to actually tell you how you can use Vanna and train your own personal rag model using any command line for any notebook environment. 

So I'm going to go back to my jupyter notebook. So here on my Jupiter notebook I'm going to start coding and show you how you can connect to your local database with your Vanna rag model and then train it. So let's get started with that. So first things first let's install the requirements which is going to be pip install W. After that what we need to do is we need to import.

W. But let's go and import it as VN. Then there is another crucial module that we need which is going to be imported from w.

Remote and the module name is now that we have imported W as well as Vanna and default we need to import a few more prerequisites like Ponda as well as the is something known as py odbc. This is going to actually handle or odbc drivers or odbc server drivers. So I'm going to go and import the last module that we need is going to be platform. 

Now this platform module actually gives you the information about the host that is hosting your database server. So with that said now let's actually go and start working. Now the first thing that we want to create is an instance of w. So I'm going to go and use wner default method here and the first parameter is the model name. So my model name is skill curve and notice this is the rag model name. After that you have to paste in your API key here. So let's go and paste my API key here and that's all. Now your Vanna instance is all set and ready. 

Next step is to connect your database with your rag model. Now for that there is another step that you have to do. So I'm going to go and show you what you need to do. First off you will have to go and check which drivers you're running. So I'm going to go with pyodbc do drivers. This method would return all the drivers that are running on my local machine. So let's go and hit enter and as you can see I'm running three different drivers but the driver that we want here is the SQL Server driver. 

So I'm going to go and copy these drivers from here. Let's go back here and create a connection. Now to create a connection all you have to do is just go with pyodbc do connect and in this method you will have to pause a few things. The first thing is the first thing is going to be the drivers that you just soft. So I'm going to go and paste in the driver name here. Now the second thing is going to be the host that is hosting the server. So I'm going to go with platform. 

Share the AI Agent Store with friends.

Node here now. This method would return the node which is running this server. And there is one more crucial detail that you need to add is the database name. So the database that I'm posting is skill curve. And last but not least we have another parameter which is trusted connection. Make sure your trusted connection is set to yes. This will ensure that you don't see any warnings related to security and that your connection is established smoothly. 

After that let's go and close this here. And let's go and run this. Once you do that now you're connected to your database and you can actually confirm that by just returning connection variable here. And as you can see your database is actually connected on this memory socket here. Now that our connections are established let's actually go to the documentation of Vanna Ai and I will show you how you can train this. 

So to train your rag model you have three types of data. The real statements, there is document strings as well as the SQL statements. But you can pair SQL statements with questions of natural language prompt to make it more impactful. And here you also get the training plan code as well. 

And first things first I'm going to create this runes skill function for myself. So let's go back to jupyter notebook and create this method. So all you have to do is you have to define a method which is going to be run esql as we saw. And then you will have to pass in a parameter which is going to be SQ. And make sure that you assign it at type which is going to be string. 

Now I'm going to go and quickly complete this method and then I will explain how it works. Now this me is really straightforward. We are just using the D frames and we are reading the SQL query from the SQL parameter that we're going to pass in here and the connection that we just created here. So I'm going to go and run this method now. The next thing that we need is to set this method using Vanna. So I'm going to go and say w. 

Share the AI Agent Store with friends.

Run esql equals to the runs method that we created as well as I want to make sure that the Run SQL flag is actually set to true. So I'm going to go and say run SQL and I'm going to say s set equals to two. 

So now that we have our run SQL method let's go back to the documentation and from here you can actually copy the code for this training plan. So I'm going to go and copy that and let's go and paste this inside my Jupiter notebook. Let's remove this command from here. 

Now as you can see we are actually creating an information schema which is going to run this method here. It is going going to select everything from all the columns of this information schema and then we are passing it to this plan. If you want to see the output all you have to do is just return the plan variable here. 

So let's go and check out the results. So there you go you get the results. These are the tables: the department table, the employee table as well as the project table inside my skill curb database. 

Now this is a warning and it actually does not affect your process but if you want to get rid of this you will have to import something known as warnings and then you will have to just set the flag of no warnings to yes. So that's just personal preferences for me it does not bother me so I'm going to leave this as it is. 

Now that we have all the information from our database and we make sure that our database is connected because it has three tables here and if I can go back to my SQL Studio you can see that my skill Cur database has the department table as well as the employee and the project table. So this means that my connection is actually successful. 

Now let's go back to our Jupiter notebook. Now one more thing that I want to mention here that you can actually set the scope of what you want to train so all you have to do is say vn. 

Please share the AI Agent Store with friends.

Train and then you have to pass in the plan which is going to be the plan variable that we just created here. So we're just passing that variable here. Let's go and start raining this on ddl statements. So I'm going to go and say BN do train. So here I'm going to say something like ddl. Let's go and actually pass in a ddl Cy here. So I'm going to go and actually close this BRAC and I'm going to paste in the query right here. 

Now let's go and run this query and as you can see it successfully added the ddl. Now we can pass in as many DL statements as we want. So I'm actually going to go and copy this again but this time I'm going to change the query. So this time I want another quy. Let me go in and paste that cury here. Let's go and run the cell and there you go this was added here. 

Now let's go and add another ddl statement here and go and do that. So all I have to do is change this statement and there you go. So you can actually go on and add as many ddl statements as you want. So I'm going to go add a few more ddl statements here. All right so I'm done training this using ddl. 

Now you can train this using documentation or SQL paired questions. So I'm actually going to show you how you can add documentation and you can actually find this over at B documentation as well. So let's go and just try the documentation training. So I'm going to go and say vn. train and this time I'm going to pass in the documentation here. 

So let's say documentation equals to. Now here you can explain what this database or this rag model is going to be about and what is your business about. So for me this is an employee database which handles the department, the projects, and the personal details of an employee. So I'm going to go and write that here and that's all you have to do. 

Now if you run this it will add this documentation for your rag model and that's not all. You can even train this using the SQL ques or SQL queries paired with natural language questions. So this time I'm going to go and say vn.

Train and this time I'm going to passing two parameters. The first one is going to be the question. So I'm going to go and pass in the question here. 

And the second one is going to be the SQL for that question. So I'm going to go and pass in the SQL as well. So let's say I want to list all the employees and that is a question. So list all employees. If someone asked this question to my model, how should it answer this question? 

So I'm going to go and enter the SQL query here. So it should actually go on and select everything. So select star from the employee table, which is going to be like this. And that's all. This is how you can actually train this on different questions. 

This one was a really basic question and you can train this on complex questions. The more you train your model, the more accurate results it will provide you. So let's go and hit enter and there you go. It actually went on and added this SQL there. 

Now I'm going to go and add a few questions really quick. So I'm going to copy this. Let's go and paste that here. So instead of this question, I'm going to paste in a new question here. 

After that, I'm going to remove this SQL query from here and I'm going to paste in the query for this specific question. Let's go and run that and there you go. So I'm going to go and train this quickly on different questions. So give me a second. 

Now that we have trained our model, we can actually start asking different questions using this module. Now you can do that by just writing vn.ask and just asking questions inside here like list all the employees. 

Or you can do one more thing, which is to use the GUI app of V AI. So I'm going to go and show you both of the ways. So to ask a question inside the command line, all you have to do is just save vn. 

Share the AI Agent Store with friends.

Ask and then you can type in your national language prompt or question and hit enter. And there you go, the results are out as well as the query. So this is the cury and here is the result. So that is how you can ask questions on the command line. 

But what if you want to use the GUI app of va? All you have to do is run the flask app and the code for that is available over at their documentation. So all you have to do is copy this snippet here and go back to your Jupiter notebook and paste in the snippet here and run it. And once you run this code snippet, you will see your app is running at Local Host 8084. So just click on this link here and it will take you to the GUI app. 

Here now if you go to the training data, you can see all the training data that you just added with the questions like list all employees. There's the quy for that and there are 58 pages of these training data. So if you go ahead, you can find the documentation as well as the ddl statement and there is even the SQL with the question find the average salary of employees in each department. 

Now you can even add the training data from just clicking here and adding the training data here. No need to go back to the command line, but I wanted to show you how you can do it in both ways. So now let's actually go go and test this out. 

So I'm going to go and select this question here which is to find all the employees hired after a specific date. It even gives me the query as well as the table output of the employes that were hired and there is the CSV file, the visual representation of when who was hired as well as there's the summary. 

Now what if I want to test out this query? So I can actually go on and copy this Cur from here. I can go to my SQL Server management studio and here if I go I' open up new query here. So I can actually paste in the query here and I can test this query out. 

So let's say I want to test this query out. So I'm going to go and paste in the query here. All I have to do is just execute this query and there you go, you get the same result, the same table that you're getting at the Vanna app right here. So this makes showes that your results are quite accurate. 

But what if I want to ask another question? So let's say I want to ask a new question here. So I'm going to go with this question here which will count the number of projects each employee is working on. So there you go, it gives you the

Employee number or the employee ID as well as the project count and the C for how it actually extracted this startup. You can see the visual representation and the CSV file can be downloaded from here. 

So this is how you can train Vanna Ai and using CH gity 3.5 llm you can train your rag Moto and get rid of all the curies and start using this hassle free tool to manage your database. 

With that said, now abso video for to today we went on and explore Vanna AI. Today we saw how we can connect to our rag model with our API and then connect to our database. We even saw how we can print this on a command line environment as well as on the GUI app. 

We asked a few test questions. Overall, I'm really impressed by Vanna and the accurate results it gives. You can feel free to try it and feed it as much data as possible to make sure that your results are way more accurate. 

I hope this video was valuable. If you found this video insightful, share your thoughts or experiences in the comments below. 

Thanks for watching and we'll see you in our next video where will continue to curb your skills with the latest deck. Till then, stay curious and keep exploring. 

Please share the AI Agent Store with friends.

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

Vanna AI Video