Skip to main content
GET
Cached GPU availability with single-flight refresh

Authorizations

Authorization
string
header
required

API key with the vk_ prefix. Create one from the dashboard.

Query Parameters

provider
string

"aws" | "vast" | "digitalocean". Absent = every probeable provider we can reach.

gpu_type
string

GPU-type filter (e.g. "H100", "A100-80GB", "MI300X-192GB"). Absent = all.

region
string

Region filter. AWS: a real region. Vast: "vast-market". DigitalOcean: a DO region slug (e.g. "tor1").

Response

200 - application/json

Availability snapshot with probe timestamp; at most one live probe runs at a time

object
string
required
refreshed
boolean
required

True when THIS request ran the probe (cache was stale and the lock was free). False = served from cache.

refreshing
boolean
required

True when another request's probe is in flight; the rows below are the previous snapshot. Poll again in a few seconds for the fresh one.

availability
object[]
required
checked_at
string<date-time> | null

When the returned rows were probed. Null only when no snapshot exists yet AND a refresh is already in flight (retry shortly).

age_seconds
integer<int64> | null

Seconds since checked_at (server clock), for staleness at a glance.