// Section 8.7 · Token

ParalleliX AI credits (the demand side)

2 min8.7Token

// 7 of 9 · token mechanics

ParalleliX AI credits.

// 8.7 · the demand side · deposit to vault · meter off-chain · settle on-chain

The demand product is ParalleliX AI: a consumer chat app (ChatGPT / Claude style) that runs open-source models on network compute. Anyone can sign up; using it costs $PRLX. This is where task demand comes from, and the source of steady-state operator funding (§8.4). Per-message cost is a function of the model and the work done.

// credit cost · per message

message_cost = model_rate × tokens_processed

Two factors

// model rate · tokens processed

// 1 · model_rate

A per-model credit rate

A small model costs less per token than a large one. The rate is calibrated so credit pricing covers operator payout plus headroom. Exact rates are a launch parameter.

// 2 · tokens_processed

Input + output tokens

The total tokens for the message, input plus output. Image and other non-text workloads price by their own unit under the same credit ledger.

Credits, not per-message gas

// 8.7.1 · deposit to vault · meter off-chain · settle to operators on-chain

Charging on-chain for every message would burn gas on each prompt and make the product unusable. Instead, a user deposits $PRLX once, then usage is metered off-chain against that credit balance.

Deposit to the on-chain vault

A user deposits $PRLX into the on-chain ParalleliXAICredits vault, an on-chain transfer they can withdraw from any time. Deposit and withdraw are the on-chain steps; spending in between is metered off-chain.

Credited to the ledger

The deposit is credited to an off-chain ledger as usage credits. From here, usage is metered off-chain.

Meter per message, off-chain

Each message is metered and debited from the credit balance off-chain. No gas, no signature per message.

Settle to operators

The project periodically settles the corresponding $PRLX to the operators whose nodes served the requests. This is the steady-state operator funding source.

Caveat·On-chain custody, off-chain meter, stated plainly

Deposits sit in an immutable on-chain vault (ParalleliXAICredits) you can withdraw from any time, and the operator share settles on-chain per served request. Spending itself is metered off-chain by the coordinator, so it is not trustless per message, but funds are never project-held and never trapped. Non-custodial per-message settlement is a planned later addition.

Open-source and verifiable

// open-source models · verified Proof-of-Execution · project-GPU fallback

ParalleliX AI runs open-source models on the operator node network. The differentiator is not raw model size; it is verifiable inference, every answer served by a real GPU node with a Proof-of-Execution, on decentralized compute rather than a central API.

If no eligible node is online, it falls back to project-operated GPUs and server CPU so the product never hard-fails. That fallback is a bootstrap measure carrying a shrinking share of traffic as the operator set grows.

Two worked examples

// launch rates illustrative · figures not final

// example A · small open-source model · short exchange

model_rate        = 0.002 PRLX / 1k tokens   (small model)
tokens_processed  = 1,200  (≈ 400 in + 800 out)
message_cost      = 0.002 × (1,200 / 1,000)   = 0.0024 PRLX

// example B · large open-source model · long context

model_rate        = 0.02 PRLX / 1k tokens    (large model)
tokens_processed  = 9,000  (≈ 6,000 in + 3,000 out)
message_cost      = 0.02 × (9,000 / 1,000)    = 0.18 PRLX

Illustrative. Model rates are a launch parameter calibrated to cover operator payout plus headroom.

// Reading path

Token Mechanics · parent hub Reward Formula · how AI usage funds operators in steady state Console · credit balance and usage in the Console