Capri Portal Settings (Legacy)

In the settings page of the Capri portal, you will have two different tabs with two different purposes. The first tab is the CRM tab, and this is where you will choose some of the GHL connection settings!

Calendar

In this page, under the calendar dropdown, you will find a list of all of the different calendars that are in the sub-account you have selected. The calendar you choose here will be the default calendar that the AI will access inside of GHL when offering times and booking and the like! You are able to have the AI use a different calendar by telling it which to use in the webhook step in GHL, though, and we'll go over that later!

Primary Objective

The main thing to worry about in the primary objective dropdown is whether or not you want the AI to be able to have access to the calendar. If so, you just need to have the primary objective set to "book appointment". Bear in mind, however, you will still need to add a data point to the webhook in GHL for this as well, which we'll cover later!

Connect Bot When

This is where you can choose when the AI will actually respond to webhooks that are being triggered in the CRM! In almost every situation, you should just have this set to "control with CRM workflows", but if you want to have certain criteria decide if the AI should respond to a contact when the webhook in the CRM is triggered, you can add those here, and they can be certain tags, the lack of certain tags, or the like, but once again, this kind of thing can be done in the CRM itself, and so having this set to "control with CRM workflows" is much easier!

Follow Up Stale Leads

This is where you can set up an action to be taken after a contact has stopped responding to your outgoing messages! You are able to handle this kind of thing through the CRM as well, such as by creating a workflow that sends an outreach message after a certain amount of time has passed, but if you'd rather use the portal to handle this, you can set a timer that counts how long it's been since the last outgoing message, an amount of times the follow up happens, and you can choose between 3 different actions. The first action is just a default message that's the same every time, and you can choose which channel it goes through and also set filters for it being sent or not. The second action is triggering a workflow that you have in your CRM, though once again you can just do this inside of the CRM itself rather than having the portal do this. The third and final action is to generate an outreach message using the AI, where you can set things like a prompt, the model, the temperature, etc. though you should just do this through an outreach webhook in the CRM rather than the portal, as you can set situational prompts and the like there, whereas in the portal, you can only have one prompt!

Moving on to the AI settings tab, this is where you will fill out the settings to be used inside of the emulator! Once these are all filled out properly, you'll able to use the emulator to simulate what a live conversation would be like so you can train the AI! Remember, though, these settings are separate from the CRM, so even though you've filled them out here, you will also need to fill them out in the CRM webhooks that you use!

OpenAI Key

This is where you will enter in the API key that you generate at https://platform.openai.com/account/api-keys, though remember, you can only ever see the key once, so make sure to write it down! While you're at the OpenAI website, be sure to enter a payment method at https://platform.openai.com/account/billing/payment-methods, as this will save you a lot of heartache and troubleshooting later!

Prompt

This second section is where you will enter the prompt that you want to use inside of the emulator! You can make this the same prompt that you're using inside of the CRM, especially if you're trying to train the AI for that prompt, or you can experiment here with prompt design to try to get the AI to better perform for your use case, such as having it ask certain questions or trying to qualify the contact! To find more information about prompts and prompt design, you can check our Prompt Design pages!

Model Type

The model type is the OpenAI model that you use inside of the emulator, and though right now we only integrate with Davinci and ChatGPT, we plan on expanding this in the future! We recommend using ChatGPT in pretty much every situation for now, however, as it is not only more powerful than Davinci, but also about 10x cheaper!

Temperature

Temperature is a measure of how "creative" the AI will get with its answers, and is a scale from 0 to 1. If you have this set too low, the AI will be very "roboty", in that it will be very clear that it's not a human talking, and the AI will also likely not even attempt to answer any questions that it doesn't have a direct answer to. On the flip side, if you have the temperature set too high, the AI will likely just make things up entirely, generating completely false information! as a result, we recommend having this set to around .08 or .11, as we've found this to work relatively well for keeping the AI grounded but still willing to try to problem solve. Feel free to experiment with different numbers using the emulator, though!

Max Output Length

Max output length is a measure of how many tokens will be sent in the response to the contact! This doesn't actually affect how long of a response the AI will generate, however, so if you have this too short, the AI's message will just get cut off in the middle. As such, if you want to shorten the Ai's actual responses, you need to tell it in the prompt to keep its messages short and concise, or something along those lines!

Knowledge Base Spreadsheet

Here is where you will enter the URL of the spreadsheet that you're wanting to use inside of the emulator. Remember, the spreadsheet needs to be formatted in that the A1 cell says "Contact" and the B1 cell says "Assistant", and the spreadsheet needs to be a basic, default Google sheet! You can find an example template that shows you how it should be set up at https://docs.google.com/spreadsheets/d/1aDcCR9mXipJNaKi45GGz_Rw-W13oQp2702FzkPVcwt0/edit#gid=0

Frequency and Presence Penalty

Frequency and presence penalty are essentially two sides of the same coin. These two settings are meant to reduce the chance of repeat phrases or words in the messages that the AI generates, and they accomplish this in two different ways! Basically, frequency penalty works by checking how many times a specific word has been said, and if it's been used repeatedly, the AI will use synonyms, and presence penalty checks how many times entire topics or phrases have come up in the conversation and tries to avoid repeating them. These two settings go from 0 to 1, and we recommend only having one of these turned up at a time!

Handoff Prompt

This section is where you can enter in your ideas for handoff prompts! What you enter in here wont have any effect on the emulator, but you can still use this as a place to brainstorm before putting it into the live CRM!

Fallback Mode

Under fallback mode you will be able to choose how the AI acts whenever it reaches a fallback. By default, the AI will remain silent and just add a tag to the contact called "fallback_reached", and you can then utilize CRM workflows to take action when this tag is added. You can, however, set the AI to respond as well when a fallback is reached. Generally it will always respond similarly, though, something along the lines of "I'm not sure what you mean by that, can you try to rephrase things?"