Resources

Roadmap

NoviQ advances by gate, not by promise. Here is what is proven today and the concrete steps between here and a permissionless, provably-private network.

What is proven now

Correctness (Gate A) and privacy obfuscation (Gate D) are proven on a CPU harness today. The WAN and cache gates connect them to a production swarm.

Correctness (Gate A) - bit-identical split vs whole model, across two processes over the sealed wire.
Privacy obfuscation (Gate D) - per-request signed permutation drops untrusted recovery from ~100% to ~0.3% with bit-identical output, on a mini model.
Live product - single-node native and browser workers, metered by the orchestrator.

Engine roadmap

Gate B - WAN

Two machines across the real internet, first correctness receipt with genuine latency.

Gate C - Cache

KV cache across the sharded pipeline for fast multi-token decode.

Gate D at scale

The privacy receipt on production models over WAN, not just the CPU mini model.

Speed and scale

Speculative decoding, asynchronous pipelining, MoE support, dynamic allocation.

Trustless verification

Stronger attestation than self-reported receipts; verifiable compute.

Contract roadmap

Settlement contract - move the revenue split on-chain so worker, treasury, and staking allocations are enforced by code.
Worker registry and slashing - on-chain identity and penalties for workers that fail their receipts.
Governance - staker control over privacy thresholds and economic parameters.

Product roadmap

Swarm worker v2 - layer-block workers join the pipeline; unlocks true sharded serving in production.
Permissionless join - anyone can contribute a GPU and be scheduled by trust tier and locality.
Live treasury and data - on-chain indexer wires the treasury dashboard and public aggregates to real numbers.

Sequencing rule

Correctness, then real WAN, then cache, then the privacy proof. The permissionless economy comes after the network can prove the output is right and the input stayed hidden - in that order, every time.