Developer documentation

GPU rental flow (current product)

  • GET /api/gpu?what=offers… — configure (template, GPU family/count, VRAM, CPU/RAM, disk, region, hours, transfers, budget) and compare eligible offers with exclusion reasons.
  • POST /api/gpu op=quote — pins offer, template and rate-card version; quotes expire after 1 hour.
  • POST /api/gpu op=accept — atomically holds capacity and creates the rental contract.
  • POST /api/gpu op=advance to=running|stopped|released — lifecycle with defined billing per state.
  • POST /api/gpu op=statement kind=final — reconciled statement (charges, fee, supplier, return).

Token inference API

Scoped API keys, /api/v1/models and /api/v1/chat/completions — the token API stays, on the same GPUs, with no double billing: each GPU-second is billed once. Under pricing model v0.2 tokens are billed × k_tok (band 0.85–1.5); cache 0.2×, batch 0.5×, priority 1.25×.

Admission & budgets

Funding confirmation precedes provisioning (demo offers simulate funding with no money movement). Budgets are ceilings; extensions need new acceptance and funding. Usage events carry unique source IDs; replays never double-charge.

Ask