We can also link multiple chains together. In other words, we can take the output from one chain and pass it in as the input for the next chain. Let's take the scenario where we have two chains.

The first chain is a chef that will come up with a unique recipe based on the main ingredient that we will provide. That recipe will then be passed into the next chain and that chain will be a food critic which will write a review of that recipe. So let's give this chain a name of chef and let's also change the prompt template.

Let's open up this window and let's enter a text like "You are an experienced chef. Create a unique recipe using the main ingredient and then in curly braces let's add a variable called ingredient like so. Let's save this.

Now let's map the input from the chat box to this ingredient variable by clicking on format prompt values. Let's remove subject and let's add a new key value called ingredient. That is what we called it in the prompt template.

Let's press enter. Now let's click on edit. Let's click on this input box and let's select the question from the chat box.

Let's test this out by saving the chat flow and now let's test this by clicking on chat. Let's expand this and let's enter a main ingredient like chicken and indeed we are getting a recipe back from our chef chain. Now let's have a look at how we can pass the output of this chain over to a critic chain.

First let's create that critic chain and typically when we build these flows we build them from left to right. So the data flows from the left which receives our user input and these variables including the user input and the output from this chain is passed along to the right side of this canvas. Let's go ahead and add the chain for our critic.

So under nodes go to chains then let's add the LLM chain. This chain expects a language model and a prompt so let's add the language model first by going to LLMs and let's add the openai node like so. Then let's connect this node to our chain.

Let's also set its credentials by clicking on this drop down and let's select the credentials that we created earlier and that is the nice thing about the way that flowwise is managing credentials is we can reuse those credentials throughout our nodes. We can leave the model name and temperature as is. Let's also add our prompt template so let's go to nodes let's go to prompts and let's add our prompt template.

Let's also attach the prompt template to the chain and let's enter our prompt template. Let's click on expand and let's enter something like you are a food critic. In fact it changes to you are a harsh and rude food critic.

Write a review about the following recipe and then in curly braces let's enter a placeholder called recipe like so. Let's save this. Now let's think about what we want to do here.

First the Chef chain will execute and it will generate a unique recipe based on the ingredient that we provide in the chat box and it is this recipe that should be injected into this prompt template in the recipe variable. So if we click on format prompt values and next to this recipe field if we click on edit and if we click on this field we only see the question and chat history still which is not what we want. So how can we get the output of this chain and inject it into this prompt.

We can do that by changing the output from this chain to output prediction and now let's connect this chain to the input of this prompt template. Let's now click on format prompt values and let's click on edit and when we click on this field we now have the output of Chef available over here. Let's now close this.

Let's name this chain as well as critic and let's save this. So let's test this. Let's clear the previous chat and let's enter an ingredient like potatoes.

Now let's see what happens. So we do get a response back but it's not great. It seems like the sentence is sort of starting halfway through and it's not ending correctly either and this comes down to the quality of the model that we're using over here.

Clearly this GPT 3. 5 instruct model only works for very simple use cases so I do want to show you a way more powerful model that will greatly improve the quality of the output but I wanted to show you the difference in quality in this video and hopefully you'll agree with me that the next model that I'm going to show you is way better to use going forward. Let's delete this node and let's add a different model.

So under add nodes instead of adding an LLM let's add a chat model and within chat models let's add the chat open AI model and this will give us access to a few powerful models like the GPT 4 and GPT 3. 5 turbo models that are currently being used by chat GPT. Let's set the credentials to flow wise tutorial.

Let's change the model name to the newer GPT 3. 5 turbo 1106 model and let's set the temperature to 0. 7.

Let's save this chat flow. Let's test it now. I'm going to clear this chat.

Let's enter potatoes again and let's have a look at the output and we can already see a massive improvement in the quality of the output. The sentence structure and everything is just absolutely perfect and one thing to note is although we have two chains in our flow the sheaf chain and the critic chain we only see the output of the final chain in our flow which is the critic chain. We do not see the recipe that is being generated by the sheaf however you can see that output from all the different chains by having a look at your console.

So in the console we can see the recipe that was generated and then the final result which was the output of the final chain in our flow which was the critic chain. So having a look at the console could be very helpful for debugging these flows.

Combining Multiple Chains (Prompt Chaining) - FlowiseAI Tutorial #3