NIRYUKTI for Node.js

An asynchronous interface to an independently implemented sparse C++ optimizer, with cancellation and portable HTML reports. No external optimization engine or runtime download is used.

Installation

npm install niryukti

Requires Node.js 18+, CMake 3.24+ and a C++20 compiler. Installation builds the included CPU engine locally. If npm blocks build scripts, review and approve the script or run node node_modules/niryukti/install.js. Linux is validated; other platforms require their own validation. CUDA requires a separately built engine.

Solve and report

const fs = require('node:fs');
const { solve, renderReport } = require('niryukti');

async function main() {
  const result = await solve('model.mps', {
    method: 'auto', device: 'auto', timeLimit: 60,
    tolerance: 1e-6, threads: 2
  });
  console.log(result.status, result.objective);
  fs.writeFileSync('report.html', renderReport(result));
}
main().catch(console.error);

solve accepts an MPS/LP/JSON path or native JSON model object. signal accepts an AbortSignal for subprocess cancellation. binary or NIRYUKTI_BINARY can select your compiled CUDA executable. Legacy VANTAGE_BINARY is accepted. Automatic selection is structural and memory-aware, not a universal speed promise. Nonoptimal statuses remain explicit; inspect accuracy and status before using a candidate. Report rendering does not certify the supplied result.

CLI

npx niryukti solve model.mps --method auto --device auto --json-out result.json
npx niryukti verify model.mps result.json
npx niryukti report result.json --output report.html

The authenticated HTTP service is provided by the Python package (pip install niryukti, then niryukti serve), not the Node CLI. See repository docs/api.md. Version 0.2.1 includes renderReport and the CLI report command.

Supported scope and license

LP, supported convex sparse QP, MILP and convex MIQP. Large singular PSD certification and advanced integer performance remain restricted. No general nonconvex global optimization. Original code is AGPL-3.0-only; commercial use and copying are permitted under its terms. LICENSE and third-party NOTICE are bundled.

Version 0.2.2

Adds checkpoint/resume for simplex, barrier and concurrent portfolios, guarded sparse singular-PSD ordering, general-integer bound conflicts and integer-lattice cuts. CUDA source builds also include continuous bound derivation replay, dual reconstruction and GPU CSR compaction. CPU distributions include the same mathematical core, but require a separate CUDA build for GPU execution.

Integer incumbent verification is distinct from replaying a full search-tree proof. Large/difficult PSD recognition, general dual conflict analysis and direct GPU barrier factorization remain restricted. No universal speedup is claimed.