Unfortunately, the Auto-GPT project is moving so fast, with so many changes every day, that I can't keep up with this.
The changes being made aren't small, they are major, e.g.:
- restructuring directories
- moving file locations
- rewriting entire sections
- adding selenium to open a desktop web browser (this one is a bit nutty since not everyone would even have a desktop os running it)
Which means I would have to rewrite my Dockerfile every time.
If someone wants to take this project over, please let me know since having this super simple way of spinning up a container is way more efficient than checking out their repo, setting up pip, etc.
This repository provides a convenient and secure solution to run Auto-GPT in a Docker container with a web-based terminal. Running Auto-GPT in a Docker container isolates it from the host system, preventing accidental damage from commands like rm -rf or apt install <whatever>. Additionally, it ensures a consistent and easy-to-maintain environment.
- Runs Auto-GPT in a Docker container for improved security and maintainability
- Provides a browser-based terminal UI using
gotty - Accessible via
http://127.0.0.1:8080, or by IP address
-
Clone the repository and navigate to the project directory
-
Copy the sample AI settings file and edit to suit your needs.
cp ai_settings.sample ai_settings.yaml -
Copy the sample .env file and edit to suit your needs, you can read about these settings on the auto-gpt repo.
cp .env.template .env -
Run the Docker container:
docker-compose up -d
To access the terminal UI via a browser, visit http://127.0.0.1:8080 or the IP address of the system it's running on.
The whole thing can also run in a kubernetes cluster.
First create the secrets: Remember to base_encode the secrets before putting them in (in secret.yaml)
echo -n 'some_secret_i_have' | base64
Omit the secrets from the .env and then run to have the variables loaded as a configmap and loaded as environment variables into the pods:
kubectl create configmap app-configs --from-env-file=.env
Then apply these files:
kubectl create -f app-autogpt-docker-kubernetes-default-networkpolicy.yaml,app-configmap.yaml,app-deployment.yaml,app-persistentvolumeclaim.yaml,app-service.yaml,app-secret.yaml
Find the internal IP of the pod
kubectl get nodes -o=wide
Access the lovely thing http://10.133.120.231:31000/ (using the internal address, you must be inside a VPC or the like to access. Else you could make it public by chaning the service to a load balancer .... I wouldn't recommend it though)
Useful commands
docker build -t autogpt .
doctl kubernetes cluster create torben-it-kubcluster --auto-upgrade=true --region ams3 --node-pool "name=torben-it-pool-1;count=2;auto-scale=true;min-nodes=1;max-nodes=3"
kubectl expose deployment app --port=8080 --target-port=8080 --name=app --type=LoadBalancer
kubectl exec --stdin --tty shell-demo -- /bin/bash
docker tag autogpt registry.digitalocean.com/torben-it-registry/autogpt
docker push registry.digitalocean.com/<your-registry-name>/my-python-app
cd docker/autogpt
docker build -t autogpt:v0.4.3-0.5 . --no-cache
docker tag autogpt:v0.4.3-0.5 registry.digitalocean.com/torben-it-registry/autogpt:v0.4.3-0.5
docker push registry.digitalocean.com/torben-it-registry/autogpt:v0.4.3-0.5
docker tag autogpt:v0.4.3-0.5 registry.digitalocean.com/torben-it-registry/autogpt:latest
docker push registry.digitalocean.com/torben-it-registry/autogpt:latest
cd ../../
kubectl apply -f app-autogpt-docker-kubernetes-default-networkpolicy.yaml,app-configmap.yaml,app-deployment.yaml,app-persistentvolumeclaim.yaml,app-service.yaml,app-secret.yaml
- Check if
gottycan pass audio from the autogpt--speakoption - Verify that integration with ElevenLabs is functional (should be okay, but untested)