Developers

API

The public routes behind the site. JSON unless noted.

A run is two requests. Everything between them happens in the browser.

Runs

Route	Params	Returns	CDN cache	Errors
`POST /api/ticket`	none	{ ticket, issuedAt }	no-store	429 rate-limited
`POST /api/runs`	Submission, JSON, 256 KB max	201 SubmitResponse	no-store	400 invalid-json, invalid-submission; 409 ticket-used; 413 too-large; 422 bad-ticket, too-fast, ticket-expired, duration-mismatch; 429 rate-limited
`GET /api/runs`	limit, 1 to 100, default 20	Newest dies, by seq	5 s, swr 30 s	none
`GET /api/runs/{id}`	run id	Run with detail: phases, traces, adapter	60 s, swr 600 s	404 not-found, also for rejected runs

Submission

{
  ticket: string,                  // from POST /api/ticket
  operatorKey: string,             // 32 to 64 hex; stored only as a 16 hex SHA-256 prefix
  result: BenchResult,             // engine output: phases, metrics, adapter, timing
  env: {
    renderer: string | null,       // WebGL renderer string
    glVendor: string | null,
    userAgent: string,             // reduced to browser and OS
    declaredFormFactor?: "laptop" | "desktop" | null,
    hiddenDuringRun: boolean
  }
}

SubmitResponse

{
  id: string,
  status: "verified" | "unranked" | "rejected",
  flags: string[],
  seq, layer, cell: number | null, // null when rejected
  score: number,
  memprint: string,
  deviceClass: { id, name, vendor, source },
  percentile: {                    // top is null below 5 runs
    class: { top, n },
    global: { top, n }
  },
  rewards: { kind, points, note }[]
}

The Stack

Route	Params	Returns	CDN cache	Errors
`GET /api/stack`	none	{ totals, recent }, 12 newest dies	5 s, swr 30 s	none
`GET /api/stack/cells`	from, a seq	application/octet-stream	10 s, swr 60 s	none
`GET /api/stack/die`	seq	{ id, seq, t, class_name, vendor, read_gbps, score, status, memprint, synthetic }	300 s, swr 1 h	400 seq required; 404 not-found

The cell stream is a 16-byte header then 8 bytes per die; The Stack has the byte layout. The Stack page polls it every 20 s with from set to the dies it already holds; the home page polls /api/stack at the same rate.

Boards

Route	Params	Returns	CDN cache	Errors
`GET /api/leaderboard`	board: class (needs class), daily (default), weekly, efficiency	Up to 50 rows, best run per operator	10 s, swr 60 s	400 class required, unknown board
`GET /api/classes`	none	Per class: runs, operators, median score and read, best, published peak	20 s, swr 120 s	none
`GET /api/rewards`	operator, 16 hex, optional	{ season, standings }; with operator also { summary, standing }	10 s, swr 60 s; no-store with operator	400 invalid operator

Token

Route	Params	Returns	CDN cache	Errors
`GET /api/token`	none	Phase, token, launch, market, holders, 24 h volume	10 s, swr 60 s	503 unavailable
`GET /api/token/trades`	limit, 1 to 100, default 30	Newest trades: side, amounts, trader, time	5 s, swr 30 s	none
`GET /api/token/chart`	range: 1h, 24h, 7d or all	OHLC candles of execution price, with volume	15 s, swr 60 s	none
`GET /api/token/holders`	none	{ count, top }, labelled protocol addresses	30 s, swr 120 s	none
`GET /api/wallet`	operator, 16 hex	{ wallet } linked to the operator, with its balance	no-store	400 bad-operator
`POST /api/wallet`	{ key, address, issuedAt, signature }	{ wallet }	no-store	400 bad input or expired; 401 signature-mismatch; 429 rate-limited

Network and claims

Route	Params	Returns	CDN cache	Errors
`GET /api/network`	none	{ totals, proof, instant, epochs, contracts, rules }: live network state, as on /network	15 s, swr 60 s	503 unavailable
`GET /api/rewards/instant`	run, a run id; or operator, 16 hex; or none	{ perRunUsd, walletDailyUsd, dailyBudgetUsd, payout \| payouts \| today }: status (pending_wallet, queued, sending, sent, failed, expired, skipped), reason, usd, amount, address, tx	no-store; 10 s without parameters	400 bad-run or bad-operator; 429 rate-limited; 503 unavailable
`GET /api/claims`	address, 0x wallet	{ contract, rwa, items }: per epoch amount, symbol, state, expiry and, when claimable, the Merkle proof	no-store	400 bad-address; 429 rate-limited; 503 unavailable
`POST /api/claims/relay`	{ address, epochId }	{ ok, tx, url, status }: an earlier daily pool claim, sent gas-free by the relayer to the wallet itself	no-store	403 region or attest; 404 not-found; 409 not-claimable, not-relayable, below-minimum; 429 daily-limit, wallet-limit, rate-limited; 503 relay-off, low-balance. Refusals carry fallback: true when the wallet can still claim itself
`POST /api/claims/attest`	{ address, issuedAt, signature }	{ attested }	no-store	400 bad input or expired; 401 signature-mismatch; 403 region; 429 rate-limited
`GET /api/proof`	none	Recent daily roots: day, leaves, root, transaction	60 s, swr 600 s	503 unavailable
`GET /api/proof/{runId}`	run id	Leaf, sibling path, day root and transaction; or { status: pending \| ineligible }	60 s; 1 h once anchored	404 not-found; 503 unavailable

A claim item's state is claimable, claimed, pending (the epoch waits for funding), expired, or attest. Stock Token items, including the daily pool, are left out for requests from the United States, and carry a proof only after the wallet has signed the attestation once. To claim, call claim(id, account, amount, proof) on the rewards contract; a weekly epoch id is uint256(keccak256(abi.encode(season, token))) and a daily one is uint256(keccak256(abi.encode("daily", dayIndex, token))), with dayIndex in whole UTC days since 1970. Items marked relay can be sent gas-free with POST /api/claims/relay.

// The message a wallet signs for POST /api/claims/attest (issuedAt within 10 minutes)
HBM: Stock Token rewards attestation

I am not a US person, and I will not claim Stock Token rewards from the United States.

Wallet: 0x…
Chain: Robinhood Chain (4663)
Issued: 2026-09-28T09:00:00.000Z

Signing is free and does not send a transaction.

Inference

Route	Params	Returns	CDN cache	Errors
`GET /api/v1/inference/estimate`	model, gpu (class id or name), quant (default q4), ctx (default 4096), batch (default 1), gpus (1 to 8), vram (GB per GPU, optional)	Fit, VRAM needed against available, decode tokens/s from measured and published bandwidth, best GPUs	60 s, swr 300 s	400 invalid-model, invalid-quant, invalid-gpu, invalid-ctx, invalid-batch, invalid-gpus, invalid-vram, each with valid values; 429 rate-limited
`GET /api/v1/inference/models`	none	Models with parameters, active parameters, layers, max context, source; weight formats	60 s, swr 300 s	429 rate-limited
`GET /api/v1/inference/gpus`	none	GPU classes with memory, published and measured bandwidth, verified runs	60 s, swr 300 s	429 rate-limited

Versioned, keyless and open to any origin (Access-Control-Allow-Origin: *). Names match loosely: gpu=4090, gpu="RTX 4090" and gpu=nvidia-rtx-4090 are the same class. Memory is in binary GB. The method is on the Inference page.

GET /api/v1/inference/estimate?model=llama-3.1-70b&quant=q4&gpu=rtx-4090

{
  "fits": false,
  "vram_needed_gb": 43.9,
  "vram_available_gb": 24,
  "memory": { "weights_gb": 39.9, "kv_cache_gb": 1.25, "overhead_gb": 2.81 },
  "decode_tok_s": { "measured": null, "spec": 22.8 },  // per stream; measured once the class has verified runs
  "batch_tok_s": 22.8,                                 // all streams, from the bandwidth named in source
  "source": "Published peak bandwidth",                // or "HBM verified runs, class median"
  "runs": 0,
  "best_gpus": ["Instinct MI300X", "H200 SXM", "H100 SXM", "A100 SXM 80GB", "H100 PCIe"],
  "best": [{ "id", "name", "vram_gb", "decode_tok_s", "source", "runs" }, …],
  "model": { … }, "quant": { … }, "gpu": { … }, "ctx": 4096, "batch": 1
}

Operations

Route	Params	Returns	CDN cache	Errors
`GET /api/health`	none	{ ok, db, ms }	no-store	503 database unavailable

Rate limits count per instance, in memory: 40 tickets and 24 submissions per hour per client address, 60 submissions per day per operator, 30 claim lookups per minute per client address, 120 inference requests per minute per client address.