The Compute Exchange • Docs

Docs

How the compute exchange works: Compute Bars, the Forge, the GPU Exchange, the agent door, and the $FOUNDRY token. Everything below runs in simulation mode until the contract deploys.

Overview

TERAHOOD is the compute exchange for Robinhood Chain — the compute layer that turns AI compute into a tradable real-world asset you can mint, hold, trade, and burn. You buy Compute Bars — prepaid on-chain credits denominated like gold ingots, where 1 gram ≈ $1 of compute — and burn them for two things: AI model answers, or raw GPU hours. Every redemption is receipted on a public proof-of-delivery ledger.

The on-chain payment is the login. No accounts, no API keys, no credit card. Your wallet is your identity, and the burn transaction is your receipt.

Compute Bars

Bars are minted from the Foundry in four denominations:

  • 1 g — retail ingot
  • 10 g — builder bar
  • 100 g — operator bar
  • 1 kg — agent-fleet kilo

Bars are transferable — hold them, gift them, or trade them. They are burned permanently on redemption, so usage directly shrinks supply. Holding $FOUNDRY unlocks mint discounts (see below).

The Forge

The Forge burns grams for AI answers. Six open frontier models sit on the menu, each at a different gram cost:

Fast answers cost ~0.05g (≈ $0.05); deep reasoning runs ~0.20g. That is 10×+ cheaper than closed-model retail, where equivalent work runs $8–15.

GPU Exchange

The GPU Exchange is the spot market for datacenter GPUs, paid in grams by the hour:

  • RTX 4090 — 0.50 g/hr
  • A100 80GB — 0.95 g/hr
  • H100 SXM — 1.40 g/hr

In simulation mode a reservation returns an instance id. The live path provisions a real instance through the GPU marketplace API and hands back SSH / Jupyter access.

The Agent Door

Robinhood Chain gave AI agents wallets. An agent with a wallet can't sign up for a closed AI provider with a credit card — so it buys its thinking here. Three doors:

OpenAI-compatible API

Point any OpenAI SDK at the Forge — zero code changes:

curl http://localhost:8791/v1/chat/completions \ -H "Authorization: Bearer FOUNDRY-SIM" \ -H "Content-Type: application/json" \ -d '{"model":"forge-8b","messages":[{"role":"user","content":"gm foundry"}]}'

x402 — pay per call

Hit the paid endpoint, receive HTTP 402 with payment terms, pay in stablecoin, retry with the receipt header:

POST /x402/forge -> 402 + payment requirements POST /x402/forge -H "X-PAYMENT: SIM:nonce" -> answer + X-PAYMENT-RESPONSE receipt

MCP — plug into any agent

claude mcp add foundry -- node compute-foundry/mcp/foundry-mcp.js tools: forge_think · forge_menu · forge_rent_gpu · forge_ledger · forge_key_balance

$FOUNDRY

$FOUNDRY is a utility token that unlocks mint-price discount tiers:

  • Hold 100K — 15% off mint
  • Hold 1M — 30% off + priority routing
  • Hold 5M — 50% off + premium model access

The Furnace market-buys $FOUNDRY with 25% of every mint's revenue and sends it to the dead address, so real usage burns supply.

Disclaimer. $FOUNDRY is utility access only — no staking, no yield, no governance, and it is not a security. Token unlaunched; all holdings on this site are simulated. Compute Bars are prepaid service credits, not an investment. Treat any circulating contract address as fake until published here.

FAQ

Is this live?

No — the whole site runs in simulation mode. Minting, redeeming, and renting update local state only. Going live needs the ComputeBar contract deployed to Robinhood Chain and an inference key on the Forge server.

Where does the AI actually run?

When the Forge server is running locally, the Forge routes to open models via wholesale inference. Otherwise an offline core answers so the full loop stays demonstrable.

Do you hold my funds?

No. TERAHOOD never custodies funds beyond the prepaid bar credit itself, never executes trades, and never gives personalized financial advice.