How to Add Custom GPTs to Instagram DMs (OpenAI GPTs Tutorial)

Liam Ottley
17 Nov 202327:53

Summary

TLDR本视频将向您展示如何快速将自定义GPT连接到Instagram,以完全自动化您的DM。通过使用GPT技术,您可以自动化客户支持和潜在客户生成等关键任务。视频提供了一个实时演示,展示了如何设置和避免常见错误。您将学习如何使用OpenAI的API、Repl托管代码、Mini Chat与Instagram集成,并处理潜在客户捕获。视频还提供了详细的步骤和模板,帮助您轻松复制和修改设置,以适应您自己的业务需求。

Takeaways

  • 🚀 本视频将展示如何快速将自定义GPT连接到Instagram,实现DMs的自动化。
  • 📈 Instagram是一个强大的销售渠道,正确使用可以大幅提升业务。
  • 🤖 连接AI助手到Instagram DMs的过程可能会非常复杂,视频提供了简化方法。
  • 🔍 视频通过实际演示,展示了如何设置自动化流程。
  • 📝 使用了特定的关键词触发自动化,如“测试”,并展示了如何与用户互动。
  • 📊 展示了如何通过GPT技术获取用户信息并进行潜在客户捕获。
  • 🛠️ 介绍了如何使用OpenAI的API和Repl托管代码来实现自动化。
  • 🔗 详细说明了如何设置和修改API密钥,以及如何与Airtable集成。
  • 🔄 解释了如何处理Mini Chat的10秒超时问题,以及如何循环检查响应。
  • 📌 提供了Mini Chat模板,用户可以根据自己的需求进行调整。
  • 📚 强调了视频教程的复杂性,并鼓励观众尝试并应用所学知识。
  • 📢 视频最后提供了资源链接,观众可以下载并使用这些资源。

Q & A

  • 视频中提到的最快的连接自定义GPT到Instagram的方法是什么?

    -视频提供了一个逐步教程,通过使用Mini Chat和Repl来自动化Instagram的DMs,从而连接GPT。

  • Instagram作为一个销售渠道有哪些优势?

    -Instagram是一个非常强大的销售渠道,当正确使用时,可以通过连接GPT来自动化关键任务,如客户支持和潜在客户生成。

  • 在Instagram DMS中连接AI助手遇到了哪些问题?

    -连接AI助手到Instagram DMS的过程可能非常痛苦,视频作者在过去几个月中遇到了持续的问题。

  • 视频中展示的自动化示例是关于什么的?

    -视频中的自动化示例是关于如何通过Instagram故事回复触发自动化,然后通过Mini Chat与GPT后端进行交互,以提供个性化的客户支持和潜在客户捕获。

  • 如何通过视频教程创建自己的GPT程序?

    -视频提供了一个模板,用户可以复制并按照步骤操作,以创建自己的GPT程序。这包括设置知识库、指令和工具。

  • 在Repl中设置GPT程序需要哪些API密钥?

    -需要OpenAI的API密钥来访问GPT模型,以及Airtable的API密钥来处理潜在客户捕获。

  • 如何确保GPT程序能够访问知识库?

    -在Repl的设置中,需要上传知识库文档,并确保在创建GPT程序时引用正确的文件名。

  • 视频教程中提到的Mini Chat的10秒超时问题是如何解决的?

    -通过将聊天端点分为两个部分:chat endpoint和check endpoint,以及在Mini Chat中设置循环检查,来解决10秒超时问题。

  • 如何将GPT程序部署到Instagram上?

    -通过在Mini Chat中设置触发器,如故事回复,然后通过Repl中的API调用与GPT程序交互,实现在Instagram上的自动化对话。

  • 视频教程中提到的AAA加速器计划是什么?

    -AAA加速器计划是一个旨在帮助个人在90天内建立盈利且可扩展的AI业务的项目。

Outlines

00:00

🚀 自动化Instagram DMs的快速方法

本视频将展示如何快速将自定义GPT连接到Instagram,以自动化DMs。Instagram是一个强大的销售渠道,通过连接GPT,可以自动化客户支持和潜在客户生成等关键任务。视频作者分享了自己过去几个月在连接AI助手到Instagram DMs时遇到的问题,并提供了一种简化的方法。

05:01

🛠️ 设置GPT自动化的步骤

视频作者详细介绍了如何设置GPT自动化,包括获取必要的API密钥,设置Airtable API密钥,以及如何在Replit上部署代码。还解释了如何通过修改代码中的特定部分来适应Instagram部署,以及如何处理Mini Chat的10秒超时问题。

10:01

🔗 连接Mini Chat并导入模板

视频展示了如何在Mini Chat上设置触发器,创建线程,并使用预设的关键词来启动自动化流程。作者提供了一个Mini Chat模板,用户可以通过简单的URL替换来自定义。同时,解释了如何通过Mini Chat与Replit后端通信,以及如何通过循环来处理超时问题。

15:02

📲 在Instagram上测试自动化流程

作者在Instagram上进行了实际测试,展示了如何通过发送特定关键词来触发自动化流程,并实时监控Replit后端的运行状态。测试包括用户输入处理、AI输出生成、以及如何通过Mini Chat与用户进行交互。

20:04

📝 捕获潜在客户信息

视频展示了如何通过自动化流程捕获潜在客户的信息,包括姓名和电话号码,并将其记录在Airtable中。作者还提到了如何通过Mini Chat的自动化流程来处理错误,并在必要时将对话转交给人工处理。

25:04

🎉 完成教程和后续步骤

视频最后,作者总结了整个教程的复杂性,并鼓励观众尝试自己设置。同时,提供了关于如何将GPT技术应用于网站和Instagram的额外信息,并邀请观众订阅频道以获取更多关于AI解决方案销售的信息。

Mindmap

Keywords

💡Instagram

Instagram是一个流行的社交媒体平台,用户可以分享图片和视频。在视频中,Instagram被用作一个强大的销售渠道,通过连接GPT技术来自动化直接消息(DMs)以进行客户支持和潜在客户生成。

💡GPTs

GPTs指的是生成预训练变换器(Generative Pre-trained Transformer)的缩写,这是一种人工智能技术,能够生成连贯的文本。在视频中,GPTs被用来创建自动化的客户服务和潜在客户生成系统。

💡自动化

自动化是指使用技术来执行通常需要人工操作的任务。视频主要讲述了如何通过GPTs技术自动化Instagram上的客户支持和潜在客户生成过程。

💡客户支持

客户支持是指企业为帮助客户解决问题和提供信息而提供的服务。在视频中,通过GPTs技术,客户支持任务如回复客户询问可以被自动化,提高效率。

💡潜在客户生成

潜在客户生成是指吸引并识别可能对产品或服务感兴趣的潜在客户的过程。视频展示了如何利用GPTs技术在Instagram上自动化这一过程。

💡直接消息(DMs)

直接消息(DMs)是Instagram上的一种功能,允许用户之间进行私密通信。在视频中,DMs被用作与客户互动和自动化沟通的渠道。

💡Mini Chat

Mini Chat在视频中指的是一个用于与用户进行交互的聊天机器人平台。它与GPTs技术结合,用于处理Instagram上的客户咨询和潜在客户信息。

💡Airtable

Airtable是一个灵活的在线数据库工具,可以用于组织和跟踪信息。在视频中,Airtable用于存储和管理通过Instagram DMs捕获的潜在客户信息。

💡API

API(应用程序编程接口)是一种允许不同软件之间进行交互的技术。视频提到了OpenAI API和Airtable API,分别用于连接GPTs技术和Airtable数据库。

💡知识库

知识库是指存储信息和数据的集合,以便人工智能系统可以访问和使用。在视频中,知识库用于GPTs技术,以提供准确的客户支持信息。

💡CRM

CRM(客户关系管理)是一种用于管理公司与客户之间互动的策略和技术。在视频中,CRM系统(如Airtable)被用于跟踪潜在客户信息,以便销售团队跟进。

Highlights

视频展示了如何将自定义GPT连接到Instagram以自动化DMs(直接消息)

Instagram是一个强大的销售渠道,正确使用可以大幅提升业务

连接AI助手到Instagram DMs可能非常困难,视频提供了一种简化的方法

视频通过实际操作演示了如何设置和避免常见错误

使用特定的关键词可以触发自动化流程

GPT能够根据用户的名字和问题提供个性化的回应

视频展示了如何通过Instagram故事自动回复并触发自动化

介绍了如何使用mini chat和repet后端进行自动化

展示了如何通过模板化代码简化GPT的创建和部署

解释了如何使用OpenAI的API来部署GPT

视频提供了如何通过mini chat模板实现Instagram的GPT部署

介绍了如何通过mini chat进行客户支持和潜在客户生成

展示了如何通过Instagram DMs捕获潜在客户信息并发送到CRM

提供了如何修改和定制GPT的知识库和提示文件

解释了如何设置OpenAI API密钥和Airtable API密钥

视频提供了如何使用mini chat模板进行Instagram自动化的详细步骤

展示了如何通过mini chat处理超过10秒的API请求超时问题

视频最后提供了如何测试和验证Instagram GPT自动化流程

Transcripts

00:00

in this video I'll be showing you step

00:01

by- step the fastest way to connect your

00:04

custom gpts to Instagram to completely

00:07

automate your DMs I'm making this video

00:09

today because Instagram can be an

00:10

incredibly powerful sales Channel when

00:12

used right and by connecting your gpts

00:14

to Instagram you can completely automate

00:17

key tasks like customer support and lead

00:19

generation for your business or for your

00:21

clients the issue is that connecting AI

00:24

assistance to Instagram DMS can be an

00:26

extremely painful process and something

00:28

that I've had constant issues with over

00:30

the past couple of months so I wanted to

00:32

create a video to walk you through the

00:34

easiest method that I've found so that

00:36

you can get things set up and avoid the

00:37

mistakes that I've made at this point in

00:39

the video I would normally give you a

00:40

big explanation of what we're building

00:42

but I think it's easier if I just show

00:43

you the finished product so we're going

00:45

to get a live demo here of exactly what

00:46

we're building in this case I'm using

00:47

the testing keyword it's going to

00:49

trigger the automation that I've set up

00:50

within mini chat which you'll see in a

00:51

second so in this case I've got some

00:53

stories up of me and Dubai if you're not

00:54

already following my Instagram it's

00:55

going to be available somewhere here you

00:56

get to see bit the behind the scenes of

00:58

my life and some of the stuff I get to

01:00

off camera um but in this case I've got

01:02

some stories up and I can reply to it of

01:05

testing and it's going to trigger the

01:09

Automation and you should see a pop up

01:11

at the top there we go so it says Hey

01:13

Liam what would you like to know about

01:14

the accelerator in this case it's

01:15

pulling the first name from my account

01:17

so if your name was Sandy or Smith that

01:19

would be using that name as the first

01:21

name um but because this account name is

01:23

Liam it's going to use Liam what would

01:24

you like to know about the accelerator

01:25

would be happy to help um who is the

01:28

owner so we can start asking some

01:31

questions in this case it's going to be

01:32

taking the question that I send there

01:33

it's going to be sending it to mini chat

01:35

and then mini chat is going to be

01:36

talking to our repet back end which

01:38

you're going to see all the code in a

01:39

second and I've templated it all so you

01:40

can just copy it and follow the steps

01:42

I'm about to show you and you'll be able

01:43

to get it up and writing so in this case

01:44

let come back Thea accelerator program

01:46

is Liam mle it's designed to help

01:48

individuals establish a profitable and

01:49

scalable AI business 90 days is there

01:51

any specific information you like to

01:52

know about the program um how how much

01:56

do the plans

01:58

cost

02:01

[Music]

02:02

the AAA accelerator offers two

02:04

membership tiers standard plan which is

02:05

$97 monthly Premium plan which is 197

02:08

which is currently sold out and so I can

02:10

keep asking questions about the program

02:12

I can ask it what are all the benefits

02:13

what do I get out of it

02:15

Etc what are the benefits on the

02:20

standard

02:22

plan

02:25

for

02:28

97

02:30

what do I

02:35

get you get access to over 15 live

02:38

workshops a month you get 24 assistant

02:40

from aam you get exclusive tools

02:41

accountability groups vets Community

02:43

templates all the good stuff so what we

02:45

can say now is to try initiate the lead

02:47

capture if I say I'd like to

02:54

join great to here I'll need your full

02:57

name and phone number including your

02:58

country code to get started so so this

02:59

is going for the lead capture now I can

03:01

say my name is um

03:06

Bruce

03:09

Wayne um plus 971 5 2 7 7

03:16

8 there we go I can send that

03:18

information it's going to pull out the

03:19

name and it's going to pull out the

03:20

phone number and it's going to send that

03:22

off to air table to log that lead so

03:24

that our sales team can follow up with

03:25

them so that's the that's the build

03:27

that's what we're going to be creating

03:28

here you can easily modify this by

03:30

changing out the knowledge base as

03:31

you'll see in the steps that are coming

03:32

in a second and you can also change the

03:34

CRM from Air table to whatever you want

03:35

but you'll be able to take the template

03:37

modified as you need and I'll be walking

03:38

you through step by step the most

03:40

valuable part of this is going to be the

03:41

mini chat template that I'm going to

03:42

give you which took me like an entire

03:44

day to create because there's so much so

03:46

many issues that I have with it due to

03:47

timeouts and stuff like this which we go

03:48

into in a sec we have a ton of stuff to

03:50

get through so let's get into it to

03:51

achieve the functionality that we just

03:52

saw we're going to need of course the

03:54

assistance API from open AI we're going

03:56

to need a repet to host that code then

03:58

we're going to call our rep code from

04:00

mini chat and I will mention it again if

04:02

you haven't seen any of my other videos

04:03

but we have the two sort of sides of

04:05

this gpts release which is firstly the

04:07

gpts on the chat GPT site where you can

04:10

create them and sort of go through them

04:11

and use them use them on the browser and

04:13

things like this and then we have the

04:14

assistance API which is really the the

04:16

meat and potatoes and how this thing is

04:18

going to be used and rolled out to

04:19

businesses and to real world

04:20

applications so while gpts are fun and

04:22

all and they're cool to create around

04:24

and sort of share around with people

04:26

most of the value is going to be created

04:27

with this assistant API so you can

04:29

deploy your gpts and your your setups

04:32

essentially to different channels in

04:33

this case we're going for an Instagram

04:35

deployment and using the GPT technology

04:37

for that on my prior videos we've done

04:39

website deployment so if you want to

04:40

check out how to put these on a website

04:42

rather than on Instagram you can do the

04:44

exact same process but put it onto a

04:45

website watching the video that's up

04:47

here I've got a couple of them now but

04:48

this is the best one to start with okay

04:49

so to get started with this build you

04:50

first need to get access to my repet as

04:52

always this is going to be available in

04:54

the first link in the description you

04:55

can go to my resource Hub sign up for

04:57

free and it's going to be at the bottom

04:58

of the list in the left panel you can

05:00

click on it there'll be a link that will

05:01

take you to this ret and you should be

05:03

greeted with a screen like this first

05:05

step is to go to the fork button on the

05:07

right we can click Fork here and we can

05:09

call it I'm just going to put a fork at

05:11

the end and then I can Fork the tripple

05:14

give it a second to boot

05:17

up then you should be greeted with

05:19

something like this for those of you who

05:20

don't know how to code don't freak out

05:21

you don't need to change any of this

05:22

really and there's only a couple parts

05:24

that I'll walk you through the process

05:25

of and that's mainly around API key so I

05:28

build this in a way that it's all ready

05:29

to go out of the box I'll just take you

05:31

through a bit of a breakdown of how this

05:32

is different from some of my other

05:33

tutorials and the way that I've had to

05:35

modify this so that it works with the

05:36

Instagram deployment this actually took

05:38

me quite a while to get working because

05:40

there's limitations of Min chat which

05:42

I'll go into later um but we can start

05:43

off where they breakdown of the main. pi

05:45

file which is where all the all the

05:47

action is happening when we're actually

05:49

interacting with this application

05:50

similar to the other builds that I've

05:51

done firstly we have the SL start

05:53

endpoint which is going to start the

05:55

conversation whenever we start a

05:56

conversation within mini chat a new

05:58

person interact with my story they hit

06:00

that trigger it's going to call this mpo

06:02

the/ start and it's going to create a

06:03

new thread we're going to give that

06:05

thread back to our mini application so

06:07

here we are giving the thread ID back

06:09

and then when the user actually asks

06:10

their question it's going to be sent to

06:12

this chat inpo and that chat inpo is

06:14

going to grab the thread ID and the

06:16

message then let's go to start a run

06:18

what's called a run in the new open AI

06:20

documentation and the way that they've

06:21

set up the assistant API so you have

06:23

what's called a run and in this

06:25

particular build we have split the chat

06:26

endpoint between this chat one and then

06:29

the second one called check now the

06:30

reason we've done that is because of the

06:32

Timeout on Mini chat is 10 seconds and

06:34

when I was testing this before I wasn't

06:36

able to get responses on Mini chat

06:38

because it was running over that

06:39

10-second timeout when I was making the

06:41

API Co so we've had to split it into

06:43

these two different parts this chat

06:44

endpoint is just going to create the run

06:46

and sort of send our information and

06:48

what we need to do up to the open AI

06:50

servers and it's going to be like okay

06:51

that's what we want now give us a couple

06:53

seconds and we'll wait for it to come

06:54

back and the way we actually grab the

06:56

response from that endpoint and we've

06:58

called the open AI API to give us our

07:00

response then we go down to this check

07:01

in point we're going to be periodically

07:03

checking in to see if our process our

07:06

our operation has been completed and if

07:08

it has been completed then we can pluck

07:09

it out and send it back to the user so

07:11

might sound a bit complicated but in

07:12

this case it's going to make a lot more

07:14

sense when we get into the mini chat

07:15

side of thing so bear with me next I'll

07:17

take you to the functions. pi file this

07:18

is where a lot of the the core

07:20

functionality is happening um in this

07:22

case we have the lead capture so this is

07:23

a air table lead capture we going to be

07:25

replacing this URL with a new one um as

07:27

you set it up on your end we have API

07:29

Keys which we'll be setting up in a

07:31

second this is the logic for creating a

07:33

new lead in the air table as I mentioned

07:35

and this is the create assistant

07:36

function which is really the main bit

07:38

that you need to wrap your head around

07:39

before going out and trying this on your

07:41

own so the create assistant function is

07:43

because when you're trying to create a

07:44

new GPT programmatically in this case

07:46

we're not using the GPT Builder we are

07:48

creating our own programmatically we

07:50

need to give the instructions or the

07:52

ingredients to the open AI API to create

07:54

our new assistant so in this case we

07:56

have of course the knowledge component

07:58

which is going to be a document here on

07:59

the left called knowledge. dox then we

08:01

have the instructions which are going to

08:03

be in the prompts dop file which we show

08:04

you in a second then we have the

08:06

functions which is set up here so I'll

08:08

show you how to change each of these

08:09

around so that you can modify this so

08:10

that you can create your own gpts for

08:12

your own use case so in this case we

08:14

have the file here which is going to be

08:16

the knowledge based document if you want

08:18

to change the knowledge base to your own

08:19

document you need to delete the

08:21

knowledge based document on the left

08:22

here and then click upload file here

08:24

upload a new file and then you need to

08:26

change the name of this file here from

08:28

knowledge. do to whatever it is in the

08:30

left panel so name here should match the

08:32

name of the document on the left hand

08:34

panel next if you want to change the

08:35

promp thing or the instruction for your

08:37

gbt it's going to be stored in this

08:39

variable here so the assistent

08:40

instructions is imported from our

08:42

prompt. piy file which is over on the

08:44

left here so if you want to change the

08:45

promp thing or the instructions for your

08:47

GPT click on the prompt stop High file

08:49

on the left I've taken this out because

08:51

it doesn't look very clean to have this

08:52

massive amount of text sort of stuffed

08:54

into your functions of P file so I've

08:56

moved it out to here so you can go in

08:58

and modify this as you like and any

08:59

changes that you make here are going to

09:00

be automatically pulled in to this

09:02

create assistant function when you

09:04

create it to to reiterate we have the

09:06

knowledge component here which we can

09:07

change we have the prompting component

09:09

here which we can change via the prompts

09:10

dop file and then finally we have the

09:13

tools section uh now the tools and order

09:15

to be able to use I should have

09:17

mentioned this in some of my other

09:18

videos so thank you for those of you who

09:19

pointed it out but the model here that

09:21

we're calling the GPT 4 preview you

09:23

actually need to have a card put on your

09:24

open AI account and you need to add some

09:26

credit to your account if you haven't

09:28

yet done that going to need that which

09:29

will show you in a sec but in order to

09:31

be able to access this new model which

09:33

is able to use GPT the much larger GPT 4

09:37

which is faster and and better in a

09:38

number of ways we need to do that step

09:40

so just to be clear some of you had

09:42

issues with that last time that is the

09:43

reason why and I'm going to be showing

09:45

you in a second how to fix that so when

09:47

it comes to tools this is the more

09:48

difficult part to change around I will

09:50

be completely honest but just to walk

09:51

you through it firstly we have the

09:53

retrieval tool here this is the

09:55

knowledge base as you can see I put a

09:56

lot of comments throughout all this

09:57

document so if you get stuck you have a

09:59

question about something not working or

10:01

you want to know how to modify things

10:03

take a look at these green comments and

10:04

you'll get a much better idea of how

10:05

this application is set up so we have

10:08

the retrieval tool which is the

10:09

knowledge base this needs to be on for

10:10

it to be able to access that knowledge

10:12

document then secondly we have just one

10:14

function here which is the lead capture

10:16

the lead capture in this case is using

10:17

the create lead function which is up

10:19

here it's referencing that name then we

10:21

get to the two parameters required to

10:23

call this function firstly we have the

10:25

name which is obviously the full name of

10:27

the lead and then we have the phone

10:28

which is the phone number of the lead

10:29

including the country code it's

10:31

important to know that to operate this

10:32

application when you have the three

10:33

different ingredients you have your

10:34

knowledge base your instructions then

10:36

these and then the tooling here as well

10:38

when you go to create an assistant when

10:40

we run this app up here for the first

10:41

time it's going to look on this left

10:43

panel and see if there's an assistant.

10:45

Json file in this case you can see that

10:46

we don't have one so it's going to

10:48

follow this and say okay if there is an

10:50

assistant. Json file let's load that but

10:53

if there is not then let's go through

10:54

the creation process and then create a

10:56

new file down here so this is our way of

10:58

of not duplicating and making a ton of

11:00

different assistants when we don't need

11:01

it you're going to be charged API usage

11:03

by how many assistants that you have up

11:05

especially if you're putting a lot of

11:06

knowledge in them you get charged by the

11:08

the amount of storage that you're taking

11:09

out with your documents as well so in

11:11

this case it's only going to create a

11:13

new assistant when there isn't one

11:15

already in the side file and this also

11:16

means if you've already run the app once

11:18

and you have a assistant. Json file on

11:21

the side panel here if you want to

11:23

modify the knowledge the prompting or

11:25

the tools then you need to delete that

11:27

file and then run the app again it will

11:29

recreate your assistant. Json file with

11:31

a new set of settings okay and with that

11:33

orientation out of the way we can get

11:34

into modifying the very few parts of the

11:37

code that we actually need to change in

11:38

order for it to work on your system

11:40

firstly we have our open AI API key

11:42

pretty straightforward one you can go to

11:43

platform. open

11:45

a.com then on the left panel here you'll

11:47

have an API key section and you can

11:49

click here and create a new secret key

11:51

it's going to pop up with a a key that

11:53

you can copy and you can copy that you

11:55

can take it back over to reflet here

11:56

then you need to go to the secret

11:58

section down the bottom left we can

12:00

delete this Google Cloud API key cuz

12:02

that's not needed for this one um but in

12:04

this case we have open AI API key you

12:06

can edit this and then you can paste

12:08

yours in and then update the secret and

12:10

as I mentioned before in order to make

12:12

sure that you get access to these newer