LinklessCompute, paid in $link
Linkless fronts the compute cost, optimizes the load across the network, and serves your answer — chat inference, image generation, and network activity alike. You access it just by holding $link in your wallet. No invoices, no per-call billing.
Holding-based access
All compute on Linkless is gated by your on-chain $link balance. A connected wallet is read at session start and continuously re-checked; your holdings provision a compute allowance — a refreshing budget of compute-units (CU) proportional to stake. Tiers (Free / Plus / Pro) map balance thresholds to allowance ceilings, max concurrency, and scheduling priority. Because entitlement derives from balance rather than spend, holders keep their compute for as long as they hold.
Fee-funded compute
Demand beyond your base allowance is funded by protocol fees generated across the $link economy — marketplace swaps, supply-side settlements, and priority lanes. A fixed share of every fee accrues to a compute treasury that procures capacity in bulk (reserved + spot) and redistributes it to active holders. Linkless fronts that cost so heavy users draw from a shared, fee-subsidized pool at amortized rates instead of paying on-demand spot prices.
Why this saves cost and compute
Aggregating demand under a single scheduler lets the network drive utilization far above what isolated users achieve:
- Batching & bin-packing — requests are continuously batched and co-scheduled to maximize GPU occupancy, collapsing idle cycles.
- Idle & spot supply — suppliers contribute underused GPUs; low-priority jobs route to the cheapest viable hardware.
- Optimal configuration — the router auto-selects model variant, quantization (
FP8 / INT4), precision, and instance type per request to hit your latency and cost target. - Cache reuse & dedup — KV-cache sharing and prompt/result deduplication eliminate recomputation across the network.
- Locality-aware routing — jobs are placed to minimize cold starts, data movement, and inter-node traffic.
Bring your own model
Holders can run their own configurations — custom checkpoints, fine-tunes, LoRA adapters, and sampling parameters — settled against the same $link allowance. Model, compute tier, temperature, and token limits are surfaced directly in chat settings; the scheduler honors your config while still placing the workload on the most efficient hardware available.
Settlement
Usage is metered in compute-units and reconciled against your allowance each epoch. Base allowance refreshes per epoch; overages are drawn from fee-funded treasury capacity or deferred to the next low-cost window. All metering is transparent and on-chain-verifiable, so cost attribution is auditable end to end.
