Create a service

Deploy a new production service with auto-scaling and load balancing

POST

/auth/v1/seed/{namespace}/services

curl --request POST \
  --url /auth/v1/seed/{namespace}/services \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: application/json' \
  --data '{"name":"llama-serving","cloud":"aws","region":"us-east-1","image":"ghcr.io/acme/llama-server:v2.0","replicas":3,"port":8080,"gpus":"A100:1","cpus":"16+","memory":"64+","health_check_path":"/v1/health","min_replicas":2,"max_replicas":8,"envs":{"MODEL_ID":"meta-llama/Meta-Llama-3.1-70B-Instruct","MAX_BATCH_SIZE":"32"},"volumes":["model-cache:/models:100"]}'

Path parameters

namespace

string

required

The namespace (user or organization) to deploy the service in.

Body parameters

name

string

required

A unique name for the service within the namespace. Must be 1-64 characters. Allowed characters: alphanumeric, hyphens, underscores.

cloud

string

required

Cloud provider to deploy on. One of `aws`, `azure`, `gcp`.

region

string

required

Cloud region to deploy in (e.g., `us-east-1`, `westus2`, `europe-west1`).

zone

string

Availability zone within the region.

replicas

integer

Number of replicas to deploy.

default: 1

image

string

Container image to deploy (e.g., `ghcr.io/acme/llama-server:v2.0`).

port

integer

Port the service listens on inside the container.

default: 80

gpus

string

GPU configuration in `GPU_TYPE:COUNT` format. For example, `"A100:4"`, `"H100:8"`.

cpus

string

CPU count. Specify an exact value like `"4"` or a minimum with `"4+"`.

memory

string

Memory in GB. Specify an exact value like `"16"` or a minimum with `"16+"`.

disk_size_gb

integer

Ephemeral disk size in gigabytes.

disk_tier

string

Disk performance tier. One of `standard`, `high`, `ultra`.

default: standard

spot

boolean

Use spot (AWS), preemptible (GCP), or low-priority (Azure) instances for reduced cost.

default: false

instance_type

string

Specific cloud instance type to use, overriding automatic selection from CPU/GPU/memory requirements.

ports

integer[]

Additional ports to expose on the service.

labels

object

Cloud provider labels/tags as key-value pairs.

envs

object

Environment variables as key-value pairs.

command

string

Override the container entrypoint command.

health_check_path

string

HTTP path for health checks. The service must return a 200 on this path.

autostop

integer

Number of idle minutes before the service is automatically stopped. Useful for development services.

min_replicas

integer

Minimum number of replicas for autoscaling. Set to `0` to enable scale-to-zero.

max_replicas

integer

Maximum number of replicas for autoscaling.

volumes

string[]

Persistent volumes in `name:/mount/path:sizeGB` format. For example, `"model-cache:/models:100"`.

isolation

string

Workload isolation level. One of `container`, `process`, `microvm`.

default: container

registry_id

string

Registry credential ID for pulling private container images.

Previous → Delete a job

Next List services →