DigitalOcean Dedicated Inference Grabber

A Node.js application that automatically provisions DigitalOcean Dedicated Inference deployments when GPU capacity becomes available. It periodically checks available Dedicated Inference capacity and creates deployments up to your desired global count.

Features

• Automatic provisioning for Dedicated Inference
• Periodic polling for current capacity
• Desired count management (global across configured regions)
• Multi-region failover by checking/trying regions in order
• Webhook notifications for provisioning events
• Docker support for worker-style deployment

Prerequisites

• DigitalOcean API token with Dedicated Inference permissions
• Node.js 22+ (for direct execution)
• Existing VPC UUID in your target region(s)

Usage

With Node.js

1. Install dependencies:

npm install

2. Run:

DO_API_TOKEN="your-do-token" npm start -- \
  --region="tor1,nyc2" \
  --model_slug="mistral/mistral-7b-instruct-v3" \
  --accelerator_slug="gpu-mi300x1-192gb" \
  --vpc_uuid="your-vpc-uuid" \
  --desired_count=1 \
  --accelerator_scale=1 \
  --accelerator_type="prefill_decode" \
  --enable_public_endpoint=true \
  --check_interval_seconds=30 \
  --name_prefix="my-di"

Optional flags:

• --hugging_face_token (for gated Hugging Face models)
• --webhook_url (notifications, including Slack incoming webhook support)

With Docker

docker build -t di-grabber .

docker run -d \
  -e DO_API_TOKEN="your-do-token" \
  -e REGION="tor1,nyc2" \
  -e MODEL_SLUG="mistral/mistral-7b-instruct-v3" \
  -e ACCELERATOR_SLUG="gpu-mi300x1-192gb" \
  -e VPC_UUID="your-vpc-uuid" \
  -e DESIRED_COUNT="1" \
  -e ACCELERATOR_SCALE="1" \
  -e ACCELERATOR_TYPE="prefill_decode" \
  -e ENABLE_PUBLIC_ENDPOINT="true" \
  -e CHECK_INTERVAL_SECONDS="30" \
  -e NAME_PREFIX="my-di" \
  -e WEBHOOK_URL="https://webhook.site/<id>" \
  --name do-di-grabber \
  di-grabber

How it works

Every polling cycle, the application:

1. Lists existing Dedicated Inference deployments in the configured regions
2. Counts deployments it manages (name prefix + model + accelerator match)
3. Checks /v2/dedicated-inferences/sizes for currently enabled region/GPU capacity
4. Attempts to create additional deployments until desired_count is met

This gives behavior similar to slug-grabber, but for Dedicated Inference capacity.

Configuration reference

Parameter	Description
DO_API_TOKEN	DigitalOcean API token (required)
region / REGION	Comma-separated regions to check in order (required)
model_slug / MODEL_SLUG	Dedicated Inference model slug (required)
accelerator_slug / ACCELERATOR_SLUG	GPU slug to request (required)
vpc_uuid / VPC_UUID	VPC UUID for deployment (required)
desired_count / DESIRED_COUNT	Global desired managed deployment count (required)
accelerator_scale / ACCELERATOR_SCALE	Accelerator scale per deployment (default: 1)
accelerator_type / ACCELERATOR_TYPE	Accelerator type (default: prefill_decode)
enable_public_endpoint / ENABLE_PUBLIC_ENDPOINT	Enable public endpoint (default: true)
hugging_face_token / HUGGING_FACE_TOKEN	Optional Hugging Face token for gated models
check_interval_seconds / CHECK_INTERVAL_SECONDS	Polling interval in seconds (default: 30)
webhook_url / WEBHOOK_URL	Optional webhook notification URL
name_prefix / NAME_PREFIX	Name prefix for managed deployments (default: di-grabber)

API references

• Dedicated Inference automation guide: https://docs.digitalocean.com/products/inference/how-to/use-dedicated-inference/#create-dedicated-inference-using-automation
• Dedicated Inference API reference: https://docs.digitalocean.com/reference/api/reference/dedicated-inference/