Track B — Inference Tokens

Compute Credits API Beta

An OpenAI-compatible inference gateway metered in credits, running at pool.linkrra.com/v1. Registration, API keys, the credit ledger and per-token metering are live today. GPU capacity comes from serving nodes contributed by the pooled fleet.

Beta — capacity-limited. Inference is answered by GPU serving nodes attached by fleet contributors. When no serving node is online, /v1/models returns 503 and completions do not execute — the gateway never fakes an answer. Want capacity, early access, or to attach your GPU? weare@linkrra.com.

1 · Get an API key

One call, no login. Register any PRL wallet address; the key is shown exactly once. The same wallet earns credits when it mines on the pool.

curl -X POST https://pool.linkrra.com/v1/register \
  -H "Content-Type: application/json" \
  -d '{"address":"<your-prl-address>"}'

Returns {"key":"sk-pool-…", …} plus the stratum endpoints for earning credits by mining. If the address already holds credits, registration requires a signed ownership challenge (POST /v1/register/challenge) so a stranger cannot claim your balance.

2 · Call the API

The gateway speaks the OpenAI chat-completions dialect — point any OpenAI-compatible client at the base URL https://pool.linkrra.com/v1.

curl https://pool.linkrra.com/v1/chat/completions \
  -H "Authorization: Bearer sk-pool-…" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen2.5-7b","messages":[{"role":"user","content":"Hello"}]}'
EndpointAuthWhat it does
POST /v1/registernoneIssue an API key for a PRL wallet address
GET /v1/modelskeyModels the online fleet actually serves right now (503 when none online)
POST /v1/chat/completionskeyOpenAI-compatible chat completion, debited per token
GET /v1/credits/balancekeyCredit balance for your wallet
GET /v1/credits/usagekeyRecent debits
GET /api/billing/plansnonePublished plan catalogue
POST /v1/nodes/registerkeyAttach a GPU serving node to the fleet (probed before acceptance)

Pricing — in credits

1 credit = 1 token, input and output alike; a minimum balance of 100 credits is required before a completion runs. There are two ways to hold credits:

Earn by mining

Point a miner at the pool with the same wallet you registered (connection guide). Credits are distributed hourly to contributors out of the inference revenue pool — so in this beta, with near-zero paid inference volume, distributions can be zero. That is by design: credits are backed by revenue, never minted from nothing.

Buy a plan

The plan catalogue (Free $0 · Developer $20 · Pro $49 · Team $149 · Enterprise $499 per month) is published at /api/billing/plans. Card checkout is not yet enabled in the beta — the Stripe billing code is deployed but not switched on. To buy credits today, email weare@linkrra.com and we will provision you manually.

Where the GPUs come from

Capacity is the pooled fleet, not a rented cloud: contributors attach GPU serving nodes with POST /v1/nodes/register, the gateway health-probes each node, and /v1/models lists only what online nodes really serve — with a per-model online-node count, so "we offer this model" and "someone is up right now to answer for it" stay distinguishable. Serving-node operators earn a share of inference revenue on top of any PRL they mine. Run an H100/H200/4090 and want in? weare@linkrra.com.