neat-pcie-application-builder

Neat PCIe Application Builder

Overview

Build native host applications against the installed Neat PCIe Model API. Treat the installed PCIe header, Python module, and packaged tutorials as the source of truth. This is a separate API from the Neat Library used inside the SDK or directly on a DevKit.

Workflow

1. Confirm that the application runs on an Ubuntu host connected to a Modalix PCIe Card.
2. Establish the installed API surface by reading references/source-of-truth.md.
3. Read references/model-lifecycle.md for every application. Choose synchronous run() or pipelined push()/pull() and preserve the required build and close lifecycle.
4. Read references/tensors-and-images.md when constructing inputs, consuming outputs, handling multiple inputs, or choosing tensor mode versus image mode.
5. Read references/model-options.md when configuring a card, queue, preprocessing, or object decode behavior.
6. Before claiming success, read references/build-and-validation.md and run the checks possible on the current host. Distinguish compile/import validation from connected-card validation.

Defaults

• In C++, include <simaai/neat/pcie/Model.h> and use the simaai::neat::pcie namespace.
• In Python, import pyneatpcie as pcie from the PCIe host Python environment.
• In tensor mode, inspect model.info() before allocating or naming model-ready inputs. In image mode, treat that information as the card-side preprocessing output contract, not the submitted image contract.
• Use run() for ordinary request/response inference. Use push() and pull() only when the application benefits from bounded pipelining.
• Use finite build and inference timeouts in applications that must fail predictably.
• Close every successfully built model on normal and error paths. Prefer a Python context manager.
• Keep generated applications runnable with explicit dependency, build, and run commands.

Boundaries

• Use pcie::Model and the supporting public types declared by Model.h only.
• Do not include Runtime.h or generate code using pcie::Runtime, ModelConfig, ModelId, RequestId, Completion, load(), try_enqueue(), or retrieve().
• Multiple models are allowed as independent Model objects assigned to distinct physical queues.
• Do not substitute the regular Neat Model, Graph, Node, or Run APIs. They are not part of the PCIe host application surface.
• Do not use PCIe implementation headers, construct raw GStreamer pipelines, or launch pcie-pipeline-builder directly.
• Do not add model compilation or Model SDK workflows. The input is an already compiled Neat model archive.
• Verify behavior against the installed release instead of guessing from memory or another Neat environment.