ai4Gauss — Gaussian Input File Generator

Describe your computational job in one sentence; get a properly formatted Gaussian input file (.gjf).

ai4Gauss is a small tkinter desktop application for people who run quantum chemistry calculations. You describe the job in plain English, an LLM writes the Gaussian route section, and the program combines it with the atomic coordinates taken from an existing .gjf file and with your server parameters (%nproc / %mem) into a complete, ready-to-submit input file.

Zero dependencies — Python standard library only. No bundled credentials — you supply your own OpenAI-compatible API key, and requests go straight from your machine to the endpoint you configured.

Installation

pip install ai4gauss

Then launch it:

ai4gauss

or, equivalently:

python -m ai4gauss

Requirements: Python 3.8 or newer with tkinter enabled. tkinter ships with the official Windows and macOS installers, but on many Linux distributions it is a separate package:

sudo apt install python3-tk        # Debian / Ubuntu
sudo dnf install python3-tkinter   # Fedora / RHEL

A standalone Windows .exe (no Python required) is available from the Releases page.

First run

Click ⚙ Open Settings… in the top-right panel and fill in:

No API key is bundled. Until you save one the status line stays red ("No API Key — open Settings to add your own ✗") and the program offers to open Settings for you at launch. Only Generate Keywords with AI needs a key — coordinate parsing and file generation work fully offline, and you can always type the route line yourself.

What it solves

1. Nobody remembers route sections — B3LYP or M06-2X? Which pseudopotential? Let the model propose one, and tell you why;
2. The trailing section is easy to get wrong — output=wfn requires the file name on its own line after the coordinate block; gen / genecp require a basis-set definition block. Miss one line and Gaussian stops with an error;
3. Copy-pasting coordinates by hand is error-prone — charge, multiplicity and 8-decimal formatting are easy to get wrong at 6 p.m.

Features

Where your files go

Because an installed package lives inside site-packages — normally not writable, and wiped by pip install -U — user data is written to the standard per-user location instead:

Set the AI4GAUSS_HOME environment variable to force a different folder. The built-in Windows .exe keeps writing next to the executable, so it stays portable.

The working directory shown in the interface defaults to the folder you started the program from, so generated files land where you expect.

Documentation and examples

The full user guide, a step-by-step quick start, a build script for making your own .exe, and two worked examples (an output=wfn job and a genecp job, both produced by this program against a real model) live in the repository:

• Quick start
• User guide
• Examples

FAQ

Does it need the internet? Only Generate Keywords with AI does. Coordinate parsing, parameter entry and file generation all work offline.

Is my API key safe? It is written in plain text to ai4Gauss_en_config.json in the folder shown above, and is only ever sent to the endpoint you configured. Keep that file to yourself.

Why does the AI suggest such an expensive level of theory? It is a suggestion, not a peer-reviewed protocol. Read the rationale in the run log, then edit the route line by hand — the extra section can be re-derived from it with Infer from route.

Windows flags the .exe as malicious. A well-known false positive for PyInstaller single-file builds; add it to your trusted list.

Known limitations

• The UI uses native tkinter widgets and has only been tested on Windows;
• Non-ASCII (e.g. Chinese) file names cause trouble for wavefunction/checkpoint files; the program falls back to ASCII names automatically;
• gen / genecp basis-set blocks are model-generated — always review them before submitting;
• This build generates one file at a time. The Chinese build additionally offers batch generation over every .gjf in the working directory, and keeps its AI prompt in an external system_prompt field; this build keeps its prompts in the source.

License

CC BY-NC 4.0 — free to use, modify and share with attribution, no commercial use. Study, research and teaching require no permission.

Note that this is not an OSI-approved licence, which is why the package reports it as a non-standard licence.

Contact

Developer: chuan437 · Email: wangchuan437@126.com

Issues, feature requests and "the keywords look wrong" reports are all welcome at https://github.com/wangchuan437/ai4Gauss-EN/issues.