Skip to main content
GET
Get the platform infrastructure / fleet burn summary (Admin only)

Authorizations

Authorization
string
header
required

JWT access token for user authentication (Authorization: Bearer <jwt_token>). Create via POST /users/me/access_tokens.

Response

Infra summary retrieved successfully

Fleet burn summary returned by the admin endpoint.

Exposes only counts and burn — never host IPs or per-host model lists.

active_hosts
integer<int64>
required

Hosts serving ≥1 model.

cost_per_gpu_hour_usd
number<double>
required

Planning rate for one allocated physical GPU-hour.

cost_per_host_usd_month
number<double>
required
daily_burn_usd
number<double>
required
fetched_at
string<date-time>
required
gpu_data_stale
boolean
required

True when current GPU allocation is unavailable or served from cache.

hourly_gpu_burn_usd
number<double>
required

Current projected fleet GPU burn per hour.

idle_hosts
integer<int64>
required

Hosts serving no models (idle capacity we still pay for).

model_gpu_allocations
object[]
required

Current allocation grouped by the model label emitted by DCGM. A GPU carrying multiple labels can appear in multiple groups; the fleet total remains separately deduplicated by physical UUID.

monthly_burn_usd
number<double>
required
stale
boolean
required

True when this is last-known / fallback data because the live fetch failed.

total_allocated_gpus
integer<int64>
required

Current distinct physical GPUs allocated to a model workload.

total_hosts
integer<int64>
required