Docs

How GACHAI works

Everything from your first deposit to calling a model with your key. Five minutes, start to finish.

What is GACHAI

GACHAI sells sealed boxes of AI credits. You deposit USDC, open boxes, and win credits you can spend on real AI models — Claude, GPT, Gemini and more — through one API key.

The mental model: real money in → credits won → AI usage out. Every box pays out at least 2× its price in credits — the average pull is ~2.2× — and every roll is provably fair.

The loop

The whole thing is three moves: crank, win, spend.

  1. 1Crank a box with USDC. Pick a machine — Bronze, Silver or Gold — and crank it. The outcome is sealed before you pay, so it can’t be changed after. Not by luck, not by us.
  2. 2Win AI tokens. Never less than 2×. Every box pays at least twice its price. 1 token = $1 of real API balance — not points, not a voucher, no expiry games.
  3. 3Spend through one key. Your tokens sit behind a single API key. Point any OpenAI-compatible app at our endpoint and every model is live — Claude, GPT, Gemini, Llama and dozens more. Metered per call, down to the token.

Connect & deposit

  1. 1Connect a wallet (MetaMask, Coinbase, Rainbow, WalletConnect) or sign in with email — your account is created automatically.
  2. 2Deposit USDC on either Base or Solana — open the deposit window from the balance pill in the nav and pick your network. It becomes your Balance, the money you open boxes with. Base credits after a brief on-chain confirmation; Solana after finality (~15s).
Two networks, one Balance

Deposit on Base or Solana — whichever chain your funds are on. Same USDC, same Balance either way (USDT is accepted too). Just pick the network in the deposit window before you send.

Balance vs Credits — two different things

Your Balance is deposited USDC: it only buys boxes. Your AI Credits are winnings from boxes: they only pay for AI usage. They never convert into each other — winnings can't buy boxes, and deposits can't buy credits directly. The only way to get credits is to win them.

Opening boxes

Pick a box on the Boxes page and hit Crank. It's paid instantly from your Balance — no wallet popup. Boxes start at $50.

You win AI Credits. The amount depends on the rarity you roll — Common, Rare, Epic or Legendary — and every box pays at least 2× its price in credits, guaranteed by the published odds tables.

Every roll is committed before you pay and verifiable after: the server seals a secret seed and publishes its hash before your payment, then reveals the seed with your result. Check any pull on the Provably fair page.

Provably fair — check us, don't trust us

Every crank is sealed beforeyou pay, so the outcome can’t be changed after — not by luck, not by us. Every crank works like this:

  1. 1Commit.Before you pay, the machine commits a sealed hash of your result — public, timestamped, immutable.
  2. 2Crank. You pay, payment confirms, the capsule drops.
  3. 3Reveal. The seed is published with your result. Anyone can recompute the hash and confirm it matched the commitment.

Not “trust us.” Check us. Every crank ever made is in the public log below with its commitment and reveal side by side — find yours, find anyone’s. (The full cryptographic detail is in Architecture.)

Verify a pull

Your API key

Every account has one key, shown on your Profile. It starts with bb-live- and works like an OpenAI API key — except it spends your won AI Credits instead of charging a card.

Reveal, copy, or rotate it from the profile. Rotating kills the old key immediately — anything still using it gets a 401.

Use your key in a tool

Spending happens in your tool, not on this site. Any app that accepts an OpenAI-compatible API works: point it at the base URL, paste your key, pick a model.

Use your key

works with any OpenAI-compatible tool

Your key works anywhere an OpenAI key does — Cursor, your own code, curl. Point the tool at the base URL below, pick a model, and your tokens cover the usage. 1 token = $1 of model usage.

Base URL · point your tool here
https://api.gachai.io
Available models
gpt-4.1-mini

Every model is vetted before it's listed.

curl https://api.gachai.io/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4.1-mini","messages":[{"role":"user","content":"Hello"}]}'

copy gives the runnable version with your real key

In Cursor (or similar tools)

  1. 1Open the tool's model settings and find "OpenAI API key" + "Override base URL" (Cursor: Settings → Models).
  2. 2Set the base URL to https://api.gachai.io
  3. 3Paste your bb-live- key as the API key.
  4. 4Enter one of the available models above as a custom model name. It behaves like OpenAI from there — your credits cover the usage.

Credits & billing

1 credit = $1 of model usage. Different models cost different amounts per token — the rate next to each model in the list above — so pricier models use credits faster.

Your credits balance updates on your profile within a few seconds of spending, even when the usage happened in an external tool. When credits run low, crank another machine.

Repackage & cash out

Coming soon

Won credits spend on AI today — but soon they won’t have to stay as compute. Repackagelets you turn the AI tokens you’ve won into something liquid, two ways:

  1. 1Convert to GACHAI— claim the platform’s native token at the current rate, then hold it or trade it on the open market. The primary path, and the one the flywheel is built around.
  2. 2Resell for USDC— sell your unused credits into the open inference market and settle the proceeds to your wallet in USDC. Capacity-limited while that market grows.

You can preview it now: on your profile, the Repackage card gives a live quote for either path as you type an amount.

Not live yet

Repackaging is coming soon — today the quotes are indicative and no credits are moved. We’ll publish the exact rates and payout rails when it ships.

The GACHAI flywheel

At token launch

GACHAI is backed by revenue, not emissions. The platform’s margin, on every box and every token of inference, is routed on-chain into continuous buy-and-burn. Volume compounds into a shrinking float.

Revenue split
75%→ buyback & burn
25% → operations & liquidity
01
Usage
Machines cranked; credits spent across 90+ models.
02
Revenue
Margin on every box and every inference call.
03
75% → buy & burn
Three-quarters of revenue, routed to market buys and burns.
04
Float contracts
Structurally deflationary — supply falls as volume rises.
05
Value accrues
Real cashflow behind a shrinking supply.
↻ and it compounds

How the 75% is enforced

  1. 1The split is codified in an on-chain router— contract logic, not treasury discretion.
  2. 2Buys execute on-DEX; burns hit a dead address — every transaction on-chain.
  3. 3Cumulative burn surfaced live: verifiable, not asserted.
Mechanism, not a promise

What’s enforced on-chain is the split — 75% of revenue, to buy-and-burn. Price tracks usage; this is structure, not a forecast. Nothing here is financial advice.

Architecture & security

Under the hood, GACHAI is three isolated systems that share nothing but the ledger: a provably-fair roll engine, a double-entry credit ledger, and an OpenAI-compatible inference gateway. Here’s how each one holds up.

Provably-fair rolls

Every roll is a commit–reveal. Before you pay, the server generates a 256-bit secret seed and publishes its SHA-256 commitment, bound to your pull. The outcome is then derived by HMAC-ing that seed with your client seed and a per-roll nonce — so neither side can bias it: you can’t grind seeds after seeing the commitment, and the house can’t swap the seed after seeing your bet. Once resolved, the seed is revealed — hash it, confirm it matches the commitment, and re-derive the outcome yourself on the provably-fair page.

Odds tables are versioned and immutable: a new schedule ships as a new version, never an edit to a live one, so the terms of a past pull can never be rewritten. Each box’s minimum payout is enforced at boot — the service refuses to start if any live table pays less than 2× its price.

Double-entry ledger

Balances live in an append-only ledger with two strictly separate buckets: fiat (deposited stablecoin) and credits(winnings). Every movement — deposit, box purchase, prize, inference spend — is one signed row, and a balance is just the sum of its rows. The buckets never convert into each other, and every write is idempotent (keyed on the on-chain event or request id), so a retried call or a re-scanned block can’t double-count.

Deposits & custody

Each account gets its own on-chain deposit address, derived hierarchically (BIP-32 on Base, ed25519 on Solana) from signing keys the running service never holds— it operates watch-only and cannot move a cent. A scanner credits Base arrivals past a confirmation threshold; Solana is verified at finality. Any 6-decimal USD stablecoin we accept — USDC, bridged USDbC, and USDT — is credited 1:1.

Watch-only by design

The online system can derive and watch every deposit address but holds no key that can spend from one. A breach of the server cannot drain deposits — the signing keys live offline.

Inference gateway

Your bb-live- key authenticates against an OpenAI-compatible gateway (/v1/chat/completions, server-sent-event streaming included). Usage is metered per token at the model’s published rate and debited from your credit bucket at settlement in a single atomic step — a request can never overspend your balance or push it negative, and inference you weren’t charged for is flagged for reconciliation.

FAQ

How do I verify a roll wasn't rigged?

Each pull publishes the server-seed SHA-256 before you pay and reveals the seed after. Hash the revealed seed to confirm it matches the commitment, then HMACit with your client seed and the roll nonce and map the digest onto the published odds table — you’ll land on the exact prize you got. The fair page does it for you, or reproduce it in a few lines of code.

What's the payout math and where's the house edge?

Every box pays at least 2× its price in credits — a hard floor enforced at boot — and the average pull is ~2.2×, with a long tail up to the jackpot. The edge is the spread between the house’s cost of serving compute and its face value, plus unspent winnings — never from shorting the box below its published odds.

Which networks and tokens can I deposit?

USDC on Base or Solana, plus bridged USDbC and USDT on the same networks — all 6-decimal USD stables, credited 1:1. Anything else, or funds sent on another chain, won’t credit (though a wrong-token arrival can be recovered by support).

Is the API really OpenAI-compatible? Does it stream?

Yes — it’s a drop-in /v1/chat/completions endpoint. Point any OpenAI SDK or tool at the base URL with your bb-live- key. SSE streaming behaves exactly as it does against OpenAI.

How are per-model rates decided?

Each model’s price is its measured serving cost times a fixed margin, refreshed as costs move. The live per-token rate sits next to every model on your profile and this page — pricier models simply draw credits faster.

I get a 401 error.

Wrong or rotated key. Copy the current key from your profile — rotating invalidates old keys instantly.

I get a 402 error.

You're out of AI credits. Crank a machine to win more — deposits alone don't create credits.

Model not found.

The model isn't on the available list. Only vetted models are served — check the list on this page or your profile (it's live).

Where's my deposit?

Give it a moment — Base needs a brief confirmation window, Solana ~15 seconds. We credit USDC or USDT on Base or Solana, 1:1. Funds sent on another network, or in another token (including native SOL or ETH), won’t auto-credit — double-check the network and token you sent.

A model that worked is now erroring.

Model providers occasionally have outages. Erroring models are flagged automatically and re-checked; try another model in the meantime — your credits work on all of them.

Can I buy credits directly?

No — credits can only be won from boxes. That's the product.