Useful work between the peaks.

Not every AI job needs the same model, the same urgency, or the same serving arrangement. OFFPEAK explores a narrower proposition: open models, authorized spare compute, and a request you can price before you run.

A deliberate place to start

The proposed P0 service is one complete loop around one model revision and one region. Keeping the initial scope small makes it possible to inspect the actual serving quality, accounting, and recovery behavior.

The spending boundary matters

A prepaid balance is custodial in this design. The wallet authorizes each request, the service reserves its maximum charge, and available unreserved funds can be withdrawn to the same wallet and network. Those boundaries should be visible at the point of use.

Records with honest limits

A signed receipt and chain commitment can establish a record’s integrity and inclusion. They cannot by themselves prove correct GPU execution, truthful metering, or the quality of an answer. The product should explain that distinction plainly.

What comes after the pilot

Streaming, tool calls, direct x402, delegated budgets, and broader supplier access depend on pilot evidence and separate design work. The initial pilot does not require $OFPK. Issuance and utility remain undecided.

Based on the supplied OFFPEAK version 1.0 product and build briefs. Read the overview ↗