Metadata-Version: 2.4
Name: desireeia
Version: 0.0.2
Summary: Python wrapper for DesireeIA — local LLM inference engine
Author: Passaro Francesco Paolo
License: Proprietary
Keywords: llm,inference,local,gguf,desireeia
Requires-Python: >=3.10
Description-Content-Type: text/markdown

# desireeia

Python wrapper for **DesireeIA**, a local LLM inference engine (GGUF
models, CPU/GPU backends). This package ships a thin ctypes binding over
the native engine, with prebuilt binaries bundled per-platform so it works
without a separate native install step.

Full documentation, model compatibility notes, and the C++/CLI side of the
project live in the main repository:
https://github.com/francescopaolopassaro/DesireeIA (see the project
README for details).

Published on PyPI since 2026-09-16: https://pypi.org/project/desireeia/

## Install

```
pip install desireeia
```

## Bundled platforms

This wheel bundles the native engine for the platforms it was built on.
Check `desireeia/lib/` in the installed package for the exact list — at
minimum `win-x64` and `linux-x64` are included. If your platform/arch
isn't bundled yet, the package will raise a clear error on import rather
than silently failing.

- **Windows**: the engine DLL and its CUDA runtime (`cudart64_13.dll`,
  for GPU acceleration) are bundled. It also needs the **Microsoft
  Visual C++ Redistributable (x64)** on the system — already present on
  most Windows machines (installed by countless other apps, .NET, GPU
  drivers, etc.), but not bundled in this package. If import fails with
  a missing-DLL error, install it from
  https://aka.ms/vs/17/release/vc_redist.x64.exe and retry.
- **Linux**: the engine `.so` only depends on `libstdc++`, `libgcc_s`
  and `glibc` — present on essentially every distribution by default,
  nothing extra to install. GPU acceleration is not packaged for Linux
  yet (CPU-only); this is a known gap, not a silent limitation.
- **macOS**: not built yet (`osx-x64`/`osx-arm64` are not bundled) —
  importing on macOS will fail with a clear error until a macOS build is
  produced.

## License

See `LICENSE` — proprietary, no modification or unauthorized integration,
no AI training/ingestion without explicit written consent from the
author.
