Deploy the Creator-AI inference worker template with OpenTofu

Stand up a CPU inference VM running Ollama, Qdrant, and a Python gateway for creator workflows using the validated OpenTofu template creator-ai-worker. Keep Ollama models small and expect higher latency than GPU hosts.

Prerequisites#

You need:

• A Quake AI account with application credentials
• OpenTofu 1.6.0 or later (installation guide)
• curl available for API checks
• OpenStack credentials sourced into the shell. See the OpenStack CLI guide.
• A copy of the creator-ai-worker template from the template reference page
• Enough project quota for one c2a.large instance, one 40 GB boot volume, one private network, and one Object Storage bucket
• SSH access to the private subnet through a bastion, or set enable_public_api = true for this deployment

Read Self-Hosted AI on Cloud Infrastructure for the platform CPU-only ceiling before you pick an Ollama model tag.

Step 1: Mint credentials and configure variables#

Mint EC2-compatible credentials and export the standard AWS variables:

openstack ec2 credentials create
export AWS_ACCESS_KEY_ID=YOUR_ACCESS_KEY
export AWS_SECRET_ACCESS_KEY=YOUR_SECRET_KEY
export AWS_ENDPOINT_URL_S3=https://object.us-east-1.rumble.cloud

Copy terraform.tfvars.example to terraform.tfvars and set:

key_name         = "YOUR_KEY_NAME"
data_bucket_name = "YOUR_PROJECT_CREATOR_AI"
s3_access_key    = "YOUR_ACCESS_KEY"
s3_secret_key    = "YOUR_SECRET_KEY"

ollama_model = "llama3.2:1b"

enable_public_api = true
api_token         = "YOUR_GATEWAY_TOKEN"

The template allocates zero floating IPs by default. Setting enable_public_api = true attaches one floating IP and opens port 8080 to admin_cidr.

Step 2: Apply the template#

From the template directory, run:

tofu init
tofu plan
tofu apply

Type yes when prompted. First boot pulls the Ollama model and starts containers. This step can take 10 or more minutes on CPU.

When the run finishes, note gateway_url and data_bucket from the outputs.

Step 3: Verify the gateway#

Call the health route:

curl -sS -H "Authorization: Bearer YOUR_GATEWAY_TOKEN" \
  "$(tofu output -raw gateway_url)/health"

A healthy stack returns JSON with a status field set to ok. If the request times out, confirm cloud-init finished on the inference instance in the Console.

List the data bucket to confirm credentials wired correctly:

aws s3 ls "s3://$(tofu output -raw data_bucket)/" --endpoint-url "$AWS_ENDPOINT_URL_S3"

An empty listing is normal immediately after apply.

Next steps#

• Creator-AI inference worker template
• Creator businesses
• Run a local LLM
• Disable enable_public_api in production and reach the gateway through a bastion or VPN

Clean up#

Empty the data bucket, then run tofu destroy from the project directory:

aws s3 rm "s3://$(tofu output -raw data_bucket)" --recursive --endpoint-url "$AWS_ENDPOINT_URL_S3"
tofu destroy