Arvica Cloud API Reference

Authentication

A verified customer administrator creates a key in the platform. Use the least-privilege scopes gpu:read, gpu:write, serverless:read, serverless:write, usage:read and billing:read. Keys are company-scoped, expire within 90 days and can be revoked at any time.

export ARVICA_API_BASE="https://cloud.arvica.ai"
export ARVICA_API_KEY="arv_live_…"

curl "$ARVICA_API_BASE/v1/gpus" \
  -H "Authorization: Bearer $ARVICA_API_KEY"

Deploy a GPU instance

1. List live GPUs with GET /v1/gpus and choose a product id.
2. Request a quote with POST /v1/quotes (gpu:write). A quote does not provision compute, reserve capacity or charge anything. It is bound to your company, expires at expiresAt and can be used for one deployment.
3. Deploy with POST /v1/deployments: the quoteToken, the same configuration, consent: true and an Idempotency-Key header. consent: true authorises the billable provisioning path.
4. Arvica re-checks the quote, exact-count stock, the locked price, your SSH key (sshKeyId, or your default saved key) and prepaid credit.
5. An eligible deployment is recorded as PENDING_PROVIDER and provisioned automatically, without manual review. Poll GET /v1/deployments/{id}.

curl --request POST "$ARVICA_API_BASE/v1/quotes" \
  --header "Authorization: Bearer $ARVICA_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{ "productId": "h100-sxm-80gb", "gpuCount": 1, "storageGB": 100, "region": "Best Available", "mode": "ON_DEMAND", "term": "On demand" }'

curl --request POST "$ARVICA_API_BASE/v1/deployments" \
  --header "Authorization: Bearer $ARVICA_API_KEY" \
  --header "Idempotency-Key: GENERATE-AND-REUSE-A-UUID" \
  --header "Content-Type: application/json" \
  --data '{
    "quoteToken": "QUOTE-TOKEN-FROM-POST-V1-QUOTES",
    "name": "Production inference",
    "productId": "h100-sxm-80gb",
    "gpuCount": 1,
    "storageGB": 100,
    "region": "Best Available",
    "mode": "ON_DEMAND",
    "term": "On demand",
    "maxRuntimeHours": 0.5,
    "maxSpendAUD": 3,
    "consent": true
  }'

Safety limits are enforced by Arvica, not by your client: maxRuntimeHours terminates the instance once it has run that long. maxSpendAUD is an AUD limit converted once at admission into your wallet's settlement currency: exactly for an AUD wallet (3 becomes 300 cents), or for a USD wallet at the reference exchange rate recorded in your quote, rounded down so it never exceeds the AUD amount. The instance is terminated once its recorded usage reaches it. Without a quoted rate for a non-AUD wallet the request is refused (SPEND_CAP_CONVERSION_UNAVAILABLE). The response shows the enforced limit in deployment.spendCap. Both limits are checked about once a minute while the instance runs. DELETE /v1/deployments/{id} terminates a deployment.

Operational boundaries

The API exposes only resources belonging to the API key's company. It does not expose provider administration, provider cost, margins, other customers, or unconfirmed inventory. Do not send private keys, passwords or raw datasets in requests.