Turn your documents into answers people trust - embed, retrieve, answer, publish.
VOXELL // GPU-NATIVE INFRASTRUCTURE
Choose the shortest path to an answer.
Two platforms, built GPU-native: a Knowledge Platform that turns your documents into answers people trust, and a Real-Time Platform for state and keys that stay resident. All running where your data lives.
State and keys resident on the GPU - local-first sync and a Redis-fast cache.
Products you can use now
Self-serve where it should be. Direct deployment support where infrastructure ownership matters.
FORGE
AVAILABLE
Forge
GPU-native vector embedding API built on a proprietary CUDA engine. The Turbo tier is FREE - forge production-grade vectors at zero cost, rate-limited, no card required. Three quality tiers (Turbo, Pro, Ultra) with 87ms median latency and zero data retention - hosted, or run the same engine yourself on bare metal or Kubernetes.
Up to 75 MTEBUp to 3x FasterHosted or self-hosted
Explore Forge →
ANSWERS
AVAILABLE
Voxell Answers
Drop in a document, ask a question, get a cited answer - not chunks, not vectors. The whole retrieval pipeline, managed, on Voxell's MTEB-leading embeddings. Free tier available, no card.
Doc → AnswerCited & GroundedFree tier, no card
Try Answers →
LUX
AVAILABLE
Voxell Lux
Self-hosted, low-latency cross-device state synchronization. Local-first reads with real-time WebSocket push, flat predictable cost, deployed on your own cloud account.
~2.5 KB Client35ns/op Rate LimiterNo MQ Plumbing
Learn more →
SPACES
AVAILABLE
Voxell Spaces
Publish a corpus as a private, hosted site your people can read and ask questions of. Signed links, a per-visitor answer budget, and an access allowlist - no app to build. Your first Space is free.
1 free SpaceSigned private linksBudgeted Ask
See Spaces →
Partnership and product research
Working systems, measured receipts, and direct access to the builders. mi-go is available for FPGA routing evaluation and licensing.
COHERENCE
DESIGN PARTNER
Coherence
GPU-native KNN retrieval database. Exact k-NN + BM25 hybrid at scale, with no ANN index to build, tune, or drift.
Learn more →
MASH
DESIGN PARTNER
MASH
Adaptive GPU+CPU sorting that understands your data. Up to 18x faster than NVIDIA CUB on real-world data, never slower.
Learn more →
ART
DESIGN PARTNER
ART
Adaptive rate limiting with zero CPU overhead. Millions of decisions per second, entirely on GPU.
Learn more →
ARC
DESIGN PARTNER
ARC
GPU-native key-value + sorted-set store. Redis verbs served by a resident megakernel - 217M ops/s at 8.8µs p50, with the CPU out of the data path.
217M ops/s8.8µs p50~730× Redis
Learn more →
MI-GO
EVALUATION + LICENSING
mi-go
GPU FPGA router whose parallel routing produces legal results by construction, independently verified - demonstrated from 4,055 to 799,403 nets on a single GPU.
799,403 netsSingle GPUIndependently verified
Learn more →
Proof before procurement.
Forge and Answers are live and self-serve, and Lux is available now via private offer. Coherence and ARC ship through the Design Partner Program, with direct founder access and a say in the roadmap.