ChatGPT Telegram Bot

Built with n8n
1

OpenAI just released its API for ChatGPT.

So naturally, I fired up n8n and made a ChatGPT telegram bot

image
image1553×387 29.1 KB

6 Likes
2

image
image750×605 67.4 KB

I saw we are having the turbo model but it gives out an error at the moment.

image

3

I did it with a HTTP node. I assume the OpenAI node hasn’t been updated yet for the new API endpoint.

Best look at the API docs and import the curl example

4 Likes
4

Version 2.18.0 was just released today with a new chat option!

image
image283×535 10.2 KB

2 Likes
5

we need the Whisper API to be added too

image
image1230×976 81.8 KB

1 Like
6

Hey @RedPacketSec,

Pop open a feature request and we can add it to the list :smiley: It looks like an interesting option to add though.

7

I updated my flow to now also take audio speech and sent it to Whisper API , this then transcribes it into text, sends it to ChatGPT, Chat GPT sends it to Eleven Labs (that can clone voices) to read the text in a natural voice and then sends it back to the telegram channel, so now you can have an audio conversatioon with chatGPT in near real time (takes about 10 seconds per reply) as n8n cant convert audio files at all with any nodes, so i have to SSH it around and pull it back in again.

Works pretty well anyway

2 Likes
8

Hey Redpacket,

Is it possible I could see how you achieved this please? I have only had n8n a week or so but its already taking over my life especially with my love for AI :grinning:

9

Yeah of course when I’m back at PC will show you my flow.

1 Like
10

image
image1824×1074 93.3 KB

I have 3 different Chat Groups, all using the same Telegram Bot, the IF checks for the telegram channel ID/Group name and filters the requests.

The bottom flow checks for a voice message/audio file and then replied with a voice reply.

2 Likes
11

Love it, Thank you, will try my best this evening to get one working :grinning:

1 Like
12

i also had to (in the chat GPT node / HTTP Node ) filter out / escape special characters like " etc otherwise it caused issues.

1 Like
13

for the different chat groups you can set the style as i like to call it for the middle flow, i initially set it to be in the style of a drunk pirate and the kids would send it messages, so you can engineer the initial prompt before someone sends a message to it, if that makes sense.

image
image1305×311 13.4 KB

14

I got most things working, Only thing I am struggling with is the audio converting, installed FFmpeg but can’t work it out, also tried api2convert but again can’t get it to work correctly either.

15

This is the command i use:

image
image793×476 12.8 KB

ffmpeg -i {{ $node["Telegram"].binary.data.fileName }} {{ $node["Telegram"].binary.data.fileName.replace('.oga','.mp3') }}

for the middle execute SSH node

image

So I upload the file to another host on my network, with ffmpeg installed , send cmd line to execute and then download the output file and delete files when flow is finished.

2 Likes
16

Thank you, will give it a go later on :crossed_fingers:

1 Like
17

Its been painful uninstalling and reinstalling hundreds of FFmpeg installations but we got there :grinning:

Just the Whisper HTTP request now.

18

All working perfectly, Thanks for the advice @RedPacketSec

1 Like
19

glad i could help :slight_smile:

1 Like
20

@RedPacketSec do you mind sharing the code for these 3 nodes?

Want to implement telegram voice transcription with whisper but can’t figure out how to store the audio file through ssh and then upload to whisper