PreTensor

Runs PreDoctor tensor workloads in a local NVIDIA CUDA host.

PreTensor connects supported PreDoctor web pages to an NVIDIA CUDA inference runtime running on your own Windows computer and, only when the user selects API mode, to the authenticated PreDoctor server API on the same web origin. The extension provides a narrow bridge for user-requested tensor, text, vision inference, and Windows slide analysis. Local AI, QC, PCA, and local voice data stay on the user's computer. In the separately selected PreDoctor API mode, the extension sends the authenticated conversation request to the same `predoctor.ai` origin after an explicit in-product consent choice. It does not send that data to advertisers, analytics providers, or unrelated origins. Features: - a guided extension, Setup EXE, and local-model readiness checklist; - an explicit Local AI power-on flow that downloads the authenticated Setup artifact when the host is missing or legacy, followed by one user-run Setup EXE; - automatic host detection after that one-time run and automatic authenticated host updates from version 0.1.6 onward; - companion-provided model catalog, model selection, and locally remembered selection; - resumable model download and integrity-verification progress; - local NVIDIA CUDA tensor and model execution; - text prefill and token-by-token decoding; - local vision encoding; - Windows-only local slide QC through an opaque source handle, fixed algorithm preset, bounded progress/cancel stream, and bounded result reads; - GPU, CUDA, and runtime readiness diagnostics; and - access restricted to the declared PreDoctor Viewer origins. Reference performance: in end-to-end local Qwen3.8 27B tests, PreTensor 0.1.3 measured approximately 4,334 prefill tokens/second with a 4,096-token prompt and 38.45 decoding tokens/second on Windows, Chrome 153, and an NVIDIA RTX PRO 6000 Blackwell Max-Q GPU. A separate 16,383-token prompt measured approximately 3,980 prefill tokens/second and 37.7 decoding tokens/second. Actual performance varies with the GPU, driver, model, context length, available VRAM, and system configuration. Requirements: - Windows 10 or Windows 11, x64; - a compatible NVIDIA GPU and driver; and - the separately installed PreTensor companion runtime. The extension does not include a model or native executable. Without the companion runtime, the popup reports that the local runtime is unavailable.