Open Source at TNG

At TNG, we do recognize the importance of free software. Many of our colleagues actively contribute to various open-source projects, both during our Techdays and in their free time. TNG also publishes and maintains several of its own open-source projects, which have either evolved from or been inspired by client projects.

Open Source Software on GitHub

Please visit our GitHub page for an overview of all our open-source projects. A selection of these is presented below.

ArchUnit

ArchUnit provides a straightforward and easily extensible framework for verifying application architecture through simple unit tests. It enables you to analyze dependencies between packages, classes, and layers, detect cyclic dependencies, and much more. To achieve this, ArchUnit translates the application's bytecode into a structure of Java classes.

ArchUnitNET

With ArchUnitNET, there is also a tool available in the .NET ecosystem that enables checking the architecture of an application via unit tests. The test rules can be defined using either the Fluent API or PlantUML diagrams. During the verification process, the assemblies are loaded at runtime, and the dependencies are made available via C# classes.

JGiven

JGiven provides a pragmatic tool for writing unit tests in a domain-specific language. By embracing Behavior-Driven Development principles, tests are naturally focused on the relevant domain logic. JGiven also generates comprehensive reports that are easy to understand for both developers and non-developers, enabling quick identification of issues.

Open Source AI Projects

Our AI-related contributions are primarily represented on Hugging Face.

DeepSeek-R1T-Chimera

With the release of DeepSeek-R1T-Chimera in April 2025, we introduced an open-weights model that added R1 reasoning to DeepSeek AI V3-0324. We applied a novel construction method: The Chimera is a child LLM, using V3's shared experts augmented with a custom merge of R1's routed experts. It is not a fine-tune or distillation, but constructed from neural network parts of both parent MoE (Mixture of Experts) models.

The Model Weights are available on Hugging Face. You can try the model merge yourself on OpenRouter.

You can find details on the construction process and analysis of Chimera model variants in one of our paper.

olmOCR

We recently created a fine-tuned Optical Character Recognition (OCR) AI model based on olmOCR to help us automate our internal document processing workflows. In our new Hugging Face article “Finetuning olmOCR to be a faithful OCR-Engine”, we discuss how we trained the Vision Language Model (VLM) to reliably transcribe the headers and footers of invoices.

The Model Weights are available on Hugging Face as well.

Other Open Source Projects

Please CLI

Please CLI is an easy-to-use LLM-powered command line tool that enables users to generate complex shell commands from natural language input. It uses the capabilities of GPT-4 and other LLM models to streamline and accelerate command line tasks. Please CLI is available for many different operating systems and shells via a wide selection of package managers such as Homebrew, apt, or Nix.

Impact Framework Webpage Plugins

The Impact Framework Webpage Plugins facilitate the estimation of carbon emissions generated by website visits. This is accomplished through the Impact Frameworks, a CLI tool for calculating the environmental impact of software applications. Unlike similar online tools, this framework ensures complete transparency and user control over all parameters. The estimations are based on the Sustainable Web Design Model implemented in co2.js. The resulting values provide a preliminary assessment of the emissions, enabling a clearer understanding of the environmental footprint associated with web activities.

Momo

Momo is a lightweight and user-friendly scheduler for Node.js that stores jobs in MongoDB. It supports both interval-based and cron-style scheduling, making task automation easy. With Momo, you can simply configure your application to "run this job every five minutes" or "run this job at 9 AM every weekday".

Virtual Office

The Virtual Office is a web application that creates a virtual representation of a physical office. It allows users to easily view current and upcoming meetings, as well as see which colleagues are present. Users can connect with each other directly via Zoom video conferences. The Virtual Office enhances collaboration in remote and hybrid work environments.