QUICKSTART.
From zero to your first inference call in under 2 minutes. This rail is OpenAI-compatible — if you already have a working OpenRouter or OpenAI integration, swap two lines and you're done.
GET AN API KEY
Sign in to Bourse and generate a key. Keys start with bs_live_ and authenticate as Authorization: Bearer. You can rotate or revoke any key from the dashboard at any time.
TOP UP CREDITS
Bourse keeps a credit balance per user, denominated in USDC. Two ways to fund:
Privy onramp turns your card into USDC in your embedded wallet, then credits your balance. $5 / $20 / $50 / $100 packs.
No prior wallet required. Best for traditional apps.
Pay USDC on Solana mainnet from any wallet. Server verifies on-chain via Solana Pay reference, balance lands seconds later.
Best for teams already running on Solana.
Both flows fund the same ledger. Calls deduct in micro-USDC at settlement time. Insufficient balance returns HTTP 402 with a link to top up.
MAKE YOUR FIRST CALL
Drop-in OpenAI-compatible. Just swap the base URL and the API key.
curl https://usebourse.xyz/api/v1/chat/completions \
-H "Authorization: Bearer $BOURSE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "google/gemma-3-27b-it",
"messages": [
{"role": "user", "content": "Hello, Bourse."}
]
}'// streaming is on by default — pass stream: false for one-shot JSON // response headers you'll see on every call: // X-Bourse-Provider "Hugging Face" | "Cerebras" | ... // X-Bourse-Model "Gemma 3 27B" // human-readable // X-Bourse-Model-Id "huggingface/google/gemma-3-27b-it"
// HOW_BILLING_WORKS
- 1. PRE-FLIGHT HOLD. On request, Bourse computes an upper-bound charge from your prompt + max_tokens and atomically holds that amount. Insufficient balance returns 402 immediately, before any upstream call.
- 2. ROUTE + FAILOVER. Your model resolves to an ordered list of provider offers (cheapest first). Forwarded to the cheapest. If that errors, next candidate, hold topped up if priced higher.
- 3. SETTLE. On stream end (or response receive), Bourse computes the actual charge from token usage and reconciles the hold. Overestimation is refunded to your balance, no manual action needed.
- 4. CREATOR FEE. A fraction of the gross (set by the model creator at listing) routes to the model's on-chain revenue vault for buyback-and-burn. The rest pays the upstream provider.
// STATUS_CODES
200OKOKInference returned. Hold reconciled to actual charge.401UNAUTHORIZEDUNAUTHORIZEDMissing or revoked Bourse key.402INSUFFICIENT_QUOTAINSUFFICIENT_QUOTABalance too low for pre-flight hold. Top up and retry.404MODEL_NOT_FOUNDMODEL_NOT_FOUNDRequested model has no active provider offer.502ALL_PROVIDERS_FAILEDALL_PROVIDERS_FAILEDEvery candidate in the failover chain errored. Hold released.