# DePIN & Compute Coordination Feed

> Pay-per-call API with GPU rental prices, availability depth and cluster health for NVIDIA H100, A100 (80GB) and RTX 4090 across io.net, Akash Network and Render Network, plus migration routes that show where to move a workload. Each paid call costs 0.02 USDC (20000 atomic units, Base) and is billed through Skyfire, Coinbase x402 v2 or AgentCash MPP.

Data is compiled ahead of time by an ingestion cycle (scrapers plus DeepSeek-V3 analytics, verified against a deterministic calculation) and served from cache, so paid calls return immediately. All prices are USD per GPU-hour.

## Paid endpoints

- [GPU prices](https://depin.rdxdata.xyz/v1/compute/gpu-prices): GET. Lowest ask per provider and GPU, cheapest provider, price spread, availability depth, cluster health score and migration routes.
- [Networks](https://depin.rdxdata.xyz/v1/compute/networks): GET. Per-network status (healthy, degraded, offline), health score and free versus total GPUs where the network publishes them.

## Free discovery endpoints

- [OpenAPI contract](https://depin.rdxdata.xyz/openapi.json): schemas, security schemes and Bazaar discovery blocks.
- [Pricing scheme](https://depin.rdxdata.xyz/v1/pricing): price, network, asset, payee wallet and accepted payment headers as JSON.
- [Health](https://depin.rdxdata.xyz/health): service, config and cache status, plus the time of the last ingestion.

## How to pay

1. Call a paid endpoint without credentials. The answer is HTTP 402 with a base64-encoded x402 v2 document in the Payment-Required header (the same JSON is in the body).
2. Pay with one method, then repeat the same request with the matching header:
   - Skyfire: send a pay or kya-pay token (ES256 JWT) in kyapay-token.
   - Coinbase x402: send a signed, single-use receipt in X-402-Payment-Receipt.
   - AgentCash MPP: send a signed, single-use receipt as Authorization: Payment <receipt>.
3. Payment terms: scheme exact, network eip155:8453 (Base), amount 20000 atomic units of USDC (0.02 USDC), asset 0x833589fCD6eDb6E08f4c7C32D4f71b54bda02913, payTo 0xA0f3bab8f298766610005F2BCcDef6964b40b5e6.

Billing rules: a Skyfire token is charged only after a successful response. Receipts are single-use and are released again if the server fails to deliver. HTTP 503 FEED_NOT_READY means no snapshot exists yet; you are not charged, retry after the Retry-After seconds.

## Response shape

gpu-prices: version, generated_at, currency (USD), unit (per GPU-hour), analytics_engine (deepseek-v3, deepseek-v3+corrected or deterministic), quotes[], analytics[], migration_routes[].

- quotes[]: gpu, provider, hourly_usd (lowest ask), median_hourly_usd, gpus_total, gpus_available, freshness (live, or stale when the source failed in the latest cycle and the quote is under 24 hours old), observed_at, source_url.
- analytics[]: gpu, cheapest_provider, cheapest_hourly_usd, highest_hourly_usd, spread_pct, providers_quoted, total_available_gpus, availability_depth (deep at 100+ free GPUs, moderate at 20+, thin below that, unknown when no capacity data exists), cluster_health_score (0-100).
- migration_routes[]: gpu, from_provider (highest ask), to_provider (lowest ask), hourly_saving_usd, saving_pct, action (migrate when the saving is 10% or more, otherwise hold), rationale.

networks: version, generated_at, analytics_engine, networks[] with provider, status, health_score, live_quotes, stale_quotes, gpus_available, gpus_total, utilization_pct, source_reachable, last_observed_at, source_url.

## Notes for agents

- A provider or GPU can be missing from quotes when its source returned nothing usable. Do not assume all nine provider and GPU combinations exist.
- Only some networks publish capacity (Akash today); gpus_total and gpus_available are null elsewhere.
- Quotes are the lowest listed ask, not a guaranteed rental price. Confirm with the provider before committing a workload.
- generated_at tells you the age of the data. Repeating a call within the same ingestion cycle returns the same data, so do not pay for it twice.