Tessero for Laravel · TesseroPHP

AI assistant

Build this with Claude Code, Cursor or Copilot

Copy a Tessero-tuned prompt for Laravel integration, grounded in 10 real API signatures , into your IDE's AI. No chatbot, just exact context.

Tessero is native numerical computing for the PHP you already write, with no call-out to a Python service. The Laravel bridge, which ships inside tessero/tessero, wires its fully verified SciPy and NumPy parity into a Laravel 12 or 13 application and is example-driven.

composer config repositories.tessero vcs https://github.com/DMJ-CV-91913/tessero
composer require tessero/tessero:dev-main
vendor/bin/tessero build
php artisan vendor:publish --tag=tessero-config   # optional
php artisan tessero:doctor

The tessero:doctor command self-diagnoses your install, reporting the active backend, kernel version, and BLAS and thread configuration so you can confirm Tessero is wired up before shipping.

Backends. Tessero is not on Packagist yet and has no tagged release, so there are no prebuilt binaries. The FFI package loads a libtessero you build with vendor/bin/tessero build; the native extension is built from a source checkout. The Laravel bridge uses the extension when it is loaded and falls back to FFI. Both backends run one C kernel and return identical numbers. If no binary matches a host, calls report as unavailable via tessero:doctor and Tessero::info() rather than returning wrong answers.

What you get

TesseroServiceProvider (auto-discovered): config, a TesseroManager singleton, runtime settings, the response macro, and tessero:doctor.
Tessero facade: array, zeros, fromBytes, memmap, load, linprog, milp, solveMdp, normal, random, info, backend.
Casts\AsNDArray: store an array in an Eloquent column and read it back as an NDArray.
Rules\NumericArray: validate request input before it becomes an array.
response()->ndarray($a): a JSON response written straight from native memory.

Eloquent casts

use Tessero\Laravel\Casts\AsNDArray;

class Forecast extends Model
{
    protected $casts = [
        'quantiles'  => AsNDArray::class,                 // JSON nested arrays
        'covariance' => AsNDArray::class . ':binary',      // exact, compact
        'features'   => AsNDArray::class . ':binary,float32',
    ];
}

A json column round-trips to 17 significant digits; binary is bit-exact (including NaN payloads and -0.0). Either format is read back regardless of the configured default, so you can switch without migrating data.

Validating input

use Tessero\Laravel\Rules\NumericArray;

$data = $request->validate([
    'prices'  => ['required', new NumericArray(ndim: 1, maxElements: 8784)],
    'weights' => ['required', new NumericArray(shape: [null, 3])],  // n x 3
]);
$prices = Tessero::array($data['prices']);

The rule checks list arrays, a rectangular shape, the dimension count or exact shape, the element cap, and that every leaf is a finite number. Always cap maxElements on public endpoints: native memory is not counted by PHP’s memory_limit.

Queues & Octane

Runtime settings, threads, memory_budget, blas_threads and epsilon, are applied at boot and re-applied at the start of every Octane request and task, and before every queued job, so a long-lived worker never carries one request’s settings into the next. Solver results from the facade are plain PHP arrays with the same keys on either backend, so your controllers and jobs do not depend on which backend is installed. Dense matrix multiplication routes through OpenBLAS at BLAS-class throughput, so the blas_threads setting directly controls heavy linear-algebra performance in your workers.

Responses

return response()->ndarray($forecast->mean(0));          // 200, application/json
return response()->ndarray($matrix, 201, ['X-Model' => 'v3']);

See the examples, the full user guide, and submit what you build to the showcase.