API reference

Build on the exchange

Two products, two references. REST + JSON · keys on the keys page.

Developer API · build compute

Rent verified GPUs by the hour with escrow, deploy workloads and templates, run jobs, manage wallet & payouts. Scoped keys carry node / jobs.

Open the Developer API reference
Data API · buy market data

Live GPU price index, history, savings, supply, demand, workloads, templates and the authenticity dataset. Metered, pay-as-you-go. Scoped key carries data.

Open the Data API reference

Two products, two references — they share no endpoint. Nothing in the Data API appears in the Developer API, and vice-versa. The Data API is gated to data-scoped keys — a node/jobs key is refused there (403). Full combined schema at /docs.

quick reference
Accounts

POST /register_user create account
GET /auth/google/login Google sign-in
POST /create_api_key scoped key

Marketplace

GET /marketplace/specs public inventory
GET /marketplace/stats live totals
POST /solve intent → placement

Run compute

POST /request_vm book + escrow
POST /create_task notebook · template · vm
POST /transcode video fan-out
POST /render Blender fan-out

Distributed · 1 job, N GPUs

POST /distributed cluster across N machines
POST /jobs/rendezvous each rank posts VPN addr
GET /jobs/{id}/hostfile MPI / torchrun hostfile
GET /jobs/{id}/cluster nodes + launch cmds

Split one job across up to 100 GPUs on different machines, wired into one torchrun/NCCL cluster over the VPN. Gang-scheduled (one rank per PC), escrowed all-or-nothing.

Use your own scheduler · another provider

Already on Slurm, MPI, Ray or Kubernetes? Don't change your stack — burst into Petabyte as an extra node pool. The cluster exports as the artifacts your launcher already reads:

mpirun --hostfile hostfile -np N …
torchrun --master_addr=<rank0> …
ray start --address=<rank0>
slurm: ResumeProgram → POST /distributed

Petabyte is another provider, not an infra change.

Wallet & payouts

GET /wallet balance + earnings
POST /wallet/methods gift · USDC · bank
POST /wallet/withdraw cash out (free / ⚡instant)

Data API · metered

GET /api/v1/data/gpu-prices price index
GET /api/v1/data/gpu-prices/history time-series
GET /api/v1/data/savings vs-cloud index
GET /api/v1/data/availability live supply
GET /api/v1/data/demand bookings, GMV, realized $/hr
GET /api/v1/data/workloads job & template mix
GET /api/v1/data/templates templates bought: jobs, buyers, GMV, models
GET /api/v1/data/market inventory summary
GET /api/v1/data/benchmarks authenticity dataset
GET /api/v1/data/usage your quota (free)

Needs a data-scoped key. A free monthly quota, then pay-as-you-go from your wallet balance. Datasets are aggregate/anonymized — no seller identity.

Try it free · no signup

Two ways to explore before you spend a cent:

1 · Dummy data, keyless. Example payloads for every dataset:

curl -s https://petabyte.market/api/v1/data/sample

2 · Real data, sandbox key. Live endpoints, free & unmetered:

curl -s https://petabyte.market/api/v1/data/gpu-prices -H "X-API-KEY: pk_sandbox_petabyte_dev"

The sandbox key pk_sandbox_petabyte_dev is read-only, never billed, and safe to publish. When you're ready for production, mint a data-scoped key and top up your wallet.

Developer API reference → Data API reference → Full schema (/docs) →