Tensor / matrix unit for the Livewire liquid-surface stack: routed GEMM and elementwise paths, optional GPU residency, and a public Python API (tensor_surface).

Version: 1.1.0
Attribution (required for all uses): © 2026 Jeremy Harford & Lilian Harford

Documentation (GitHub Pages): https://lillavee.github.io/Tensor-Liquid-Surface/
Also: https://lillavee.github.io/Tensor-Liquid-Surface/TENSOR_DOCUMENTATION.html
In-tree HTML: DOCS/TENSOR_DOCUMENTATION.html

• Linux installer install/install_tensor_surface.sh (absolute --root by default; --allow-relative-paths opt-in).
• ternary_alu Rust compile sources beside the Windows wheel (owned/ternary_alu/src, Cargo/maturin path).
• cuda.rs — Windows nvcuda + Linux libcuda / dlopen (WSL-aware).
• Beginner CLI: tensor / tls / TENSOR.cmd (menu, smoke, status, demo, docs, bench, calc).
• Manual updates: Charge Windows/Linux Raybeam names with shared atomizer command; BUILD/install path-law honesty.
• CI: ship/** branch pushes run ci + linux-build; Linux NumPy parity test restored.
• Linux manylinux wheels checked in (tu_gemm + ternary_alu); installer prefers them in-tree.
• Atomizer source nest Livewire Atomizer Liquid Surface/ (open-source; build with cargo).
• Note: 2026-07-20 malfunctioning agent under Agent OS — damage contained; some lab artifacts were late to promote — see PATCH_NOTES.md.

Full notes: PATCH_NOTES.md.

One-click production install (Windows): double-click INSTALL.cmd at the project root (or run install/install_tensor_surface.cmd).

That copies into %LOCALAPPDATA%\Wet\TensorSurface\, sets up a venv, installs dependencies (NumPy, tu_gemm, owned ternary_alu, the tensor_surface API), registers full-h2h on your user PATH, and runs a tiny API smoke matmul. The development clone is left as-is.

Linux: see install/install_tensor_surface.sh and install/README.md.

Dev API: from this repo root:

pip install -e API

Put this root on PYTHONPATH, or set WET_TENSOR_ROOT. Marker file: .tensor_surface_root.

Native wheel filenames still say 1.0.0; API/docs version is 1.1.0.

Other platforms need a source build (or additional wheels). GPU seats need a host NVIDIA driver (not redistributed).

After install (or from this tree with the API on your Python path):

tensor              menu (also: TENSOR.cmd / tls)
tensor smoke        check that it works
tensor status       version, paths, GPU/CPU
tensor demo         quick demo matmul
tensor docs         open the HTML manual
tensor bench --quick
tensor calc         interactive matrix calculator
tensor calc --size 128
tensor calc --a A.npy --b B.npy --out C.npy

Same entry points: python -m tensor_surface … and tls ….

import numpy as np
import tensor_surface as ts

print(ts.__version__)  # 1.1.0
tu = ts.boot()
a = np.random.randn(256, 256).astype(np.float32)
b = np.random.randn(256, 256).astype(np.float32)
c = tu.matmul(a, b)
print(tu.status())
tu.close()

• Build notes: BUILD.md
• Agent/root notes: AGENTS.md
• Patch notes: PATCH_NOTES.md

This tree includes mirrors of:

• Livewire Vector Liquid Surface/ (nested VPU bind target)
• Livewire Arithmetic Surface/
• Livefire Floating Surface/
• Livewire Atomizer Liquid Surface/ — Atomizer / Raybeam source (atomizer/ crate)

• ALU: owned/ternary_alu/ (src/lib.rs, Cargo/maturin) + Windows/Linux wheels under wheels/
• Atomizer: Livewire Atomizer Liquid Surface/atomizer/ (shared bin atomizer; stage Raybeam names via scripts/stage_raybeam_bins.*)

Harford Tensor Surface License — see LICENSE and NOTICE. Third-party notices live under licenses/third_party/. NVIDIA drivers/CUDA are not redistributed.

• GitHub Actions CI (Windows + CPython 3.12): .github/workflows/ci.yml — also runs on ship/** pushes; installs wheels + API and runs API/tests/test_smoke_matmul.py.
• Linux build: .github/workflows/linux-build.yml — ubuntu-latest wheel + NumPy parity (tests/test_linux_parity.py); also on ship/**.
• GitHub Pages: .github/workflows/pages.yml — publishes the current manual to https://lillavee.github.io/Tensor-Liquid-Surface/ (tracks main on the public Tensor repo).
• Release tag: create v1.1.0 when you cut the release (not automatic).