KITT is a ChatGPT-powered AI that lives in a WebRTC conference call.
You can try an online demo right now at https://livekit.io/kitt
This repo contains two services:
- meet
- lkgpt-service
The meet
service is a NextJS app that implements a typical video call app. The lkgpt-service
implements KITT. When a room is created, a webhook calls a handler in lkgpt-service
which adds a participant to the room. The particpant uses GCP's speech-to-text, ChatGPT, and GCP's text-to-speech to create KITT.
The following diagram illustrates this:
GOOGLE_APPLICATION_CREDENTIALS
json body from a Google cloud account. See https://cloud.google.com/docs/authentication/application-default-credentials#GAC- OpenAI api key
- LiveKit api_key, api_secret, and url from LiveKit Cloud
- Go 1.19+ and Node.js
To run locally, you'll need to run the two services in this repo: meet
and lkgt-service
.
In the meet/
directory, copy .env.example
to .env.local
and fill in your LiveKit connection details. Then run:
yarn install && yarn dev
In the lkgpt-service/
directory, copy config-sample.yaml
to config.yaml
and fill in your LiveKit connection details and OpenAI API key. Then run:
go run /cmd/server/main.go --config config.yaml --gcp-credentials-path gcp-credentials.json
Once both services are running you can navigate to http://localhost:3000. There's one more step needed when running locally. When deployed, KITT is spawned via a LiveKit webhook, but locally - the webhook will have no way of reaching your local lkgpt-service
that's running. So you'll have to manually call an API to spawn KITT, using room_name
from the url slug when you enter a room in the Meet UI.
curl -XPOST http://localhost:3001/join/<room_name>