Private AI, paid on Fuel.
Fuel Computer is an AI chat and OpenAI-compatible API funded with FUEL or USDM on Fuel Ignition. The public service is currently a staging implementation with wallet login, API keys and prepaid accounting; testnet FUEL invoices enabled for end-to-end testing; mainnet payment acceptance remains disabled until production treasury and pricing configuration is independently verified.
API quick start
Base URL: https://api.fuel.computer/v1
curl https://api.fuel.computer/v1/chat/completions \
-H "Authorization: Bearer fuel_…" \
-H "Content-Type: application/json" \
-d '{"model":"fuel-fast","messages":[{"role":"user","content":"Hello"}]}'V1 supports text Chat Completions. Trial requests may stream; authenticated paid streaming remains disabled until terminal token usage can be settled safely. Unsupported fields are rejected rather than silently reinterpreted.
Payments
After a small trial allowance, requests require prepaid credit. Staging issues a time-limited Fuel Sepolia invoice for the verified testnet FUEL asset and credits only a transaction matching its payer, receiver, exact asset and unique amount. Testnet assets have no monetary value. Get test assets at the official Fuel faucet. FUEL transfers also require test ETH, the network base asset, for fees. Mainnet will later quote FUEL or USDM with explicit pricing and finality policy.
Payments are public on-chain and prepaid service credit is custodial after receipt.
Privacy
Fuel Computer does not write prompts or completions to its application databases, analytics, or queues. This staging build calls Workers AI directly and does not use AI Gateway; provider retention must be independently verified before production privacy claims are expanded. Chat history is stored only in your browser and can be cleared at any time. Minimal accounting metadata—model, token counts, charge, status and timestamp—is retained. Fuel Computer is private, not anonymous.
Errors
401- Invalid or revoked API key.
402- Insufficient credit. The response contains a funding URL.
413- Input exceeds the service limit.
429- Rate limit reached.
503- Inference or payments are not configured.
Pi provider
The staging @fuel-computer/pi-provider package skeleton registers this API as an OpenAI-compatible provider. It uses FUEL_COMPUTER_API_KEY, never a wallet private key, and presents funding instructions when the API returns HTTP 402.