idw – Informationsdienst Wissenschaft
Nachrichten, Termine, Experten
Nachrichten, Termine, Experten
idw - Informationsdienst
Wissenschaft
Many software developers use large language models (LLMs) in their daily work. While LLMs are useful, the ability to run the generated code or edit files could enhance productivity. Such tools, called agentic coding assistants (or AI agents) do exist. However, they have limitations, particularly in terms of safety and security, which hinder their widespread adoption in academia and industry. A team at the Center for Advanced Systems Understanding at HZDR has developed a framework that provides a big step forward for a responsible way to explore what AI agents can do: Terok offers a unique combination of features and security measures and works with both commercial and self-hosted LLMs.
AI-assisted coding is a novel, exciting and disruptive technology that was quickly adopted by software developers in both academia and industry. It offers a “chatbot mode” where the programmer has a conversation with an LLM. Users need to formulate so-called prompts to get the desired code from the LLM. This code is then transferred from the chatbot to the software project. Agentic coding assistants are even more powerful. They can generate code but also read files, execute code, and react to what happens during those steps. This is similar to the approach a human programmer would take. Developers working with these coding assistants report minimal required user interaction as the AI agents work in an autonomous and self-correcting way.
“There are ongoing discussions among programmers worldwide how to deal with agentic coding assistants,” says Dr. Andreas Knüpfer, leader of the CASUS Scientific Computing Core (SCC) team. “The opinions range from ‘it is purely evil and no programmer should touch it’ to ‘all programmers will be replaced by AI tomorrow’. We disagree with both extremes.” The SCC team saw the huge potential for programmers in general and scientific coding in particular. However, using previously available tools with real use cases from science was not an option. Agentic AI systems are most powerful when the agent is running unrestricted, but then they cannot offer protection against intentional harm or malfunctions. In the former case, sensitive data like passwords may be stolen. In the latter case, the agents may delete important data. Terok addresses a long list of potential vulnerabilities all of which have been actively exploited already.
Working with real data from relevant projects
“To explore the full potential of these new AI agent mode, we must be able to use real data from relevant projects,” says lead developer Dr. Jiří Vyskočil. “We wanted to be able to compare different LLMs and software tools, their suitability for different kinds of programming tasks, their potentials and pitfalls. So we started our work on Terok in late 2025.”
The open-source framework Terok uses containerization and a number of additional safety and security features to control how agents act on local hardware. Terok itself is not an AI agent. Rather, it allows users to switch between different LLMs they have access to. “Whatever LLM or agent you use, Terok can provide protection,” adds Vyskočil. Out of the box, it supports leading commercial agents like Claude by US company Anthropic as well as ones from the academic sector in Germany like Blablador and KISSKI Chat AI. Of course, Terok can be connected to self-hosted LLMs. It also allows to manage multiple projects with concurrent tasks in each project and provides a ready-made environment with easy installation.
“Terok can be seen as a safety and security net for agentic AI,” says Knüpfer. “Being a young academic project in a rapidly changing area, it is certainly not fully evolved and complete. But it makes a big difference compared to the naive usage of agentic LLMs on your own laptop without any such net. Users can now finally learn how all those different LLMs compare in ability and also in costs for specific tasks. To us in the team, Terok is like a key to a fun playground that has been off-limits until now.”
CASUS Director Prof. Thomas D. Kühne is an active Terok user. He encourages his team to try out this new tool: “For various reasons, I’ve been skeptical of the alternatives available so far, including OpenClaw. Now that I’ve been working with Terok for a few weeks, I have to admit that I’m totally hooked. Tasks that used to take weeks to complete can now be finished in half a day.” The SCC team is very much looking forward to hearing from the community and welcomes anyone who would like to help with further development of Terok.
Dr. Andreas Knüpfer
Research Team Leader
Center for Advanced Systems Understanding (CASUS) at HZDR
a.knuepfer@hzdr.de
https://www.casus.science/?page_id=20994
Heat equation calculation project tackled with Terok.
Source: A. Knüpfer/CASUS
Criteria of this press release:
Business and commerce, Journalists, Scientists and scholars, Students, all interested persons
Chemistry, Information technology, Materials sciences, Physics / astronomy
transregional, national
Research results, Transfer of Science or Research
English
You can combine search terms with and, or and/or not, e.g. Philo not logy.
You can use brackets to separate combinations from each other, e.g. (Philo not logy) or (Psycho and logy).
Coherent groups of words will be located as complete phrases if you put them into quotation marks, e.g. “Federal Republic of Germany”.
You can also use the advanced search without entering search terms. It will then follow the criteria you have selected (e.g. country or subject area).
If you have not selected any criteria in a given category, the entire category will be searched (e.g. all subject areas or all countries).
