Menu
The BMS Lab
The BMS Lab
UTFacultiesBMSThe BMS LabInfohubFramework PCs

Framework PCs

Labs are in high demand. Only book when necessary, with maximum of 3 consecutive days at a time. 

We are currently refining the process and guidelines for accessing the Framework PCs.

What are the Framework PCs?

The BMS Lab now offers six high-performance Framework PCs to support and accelerate AI research.

These Framework PCs can support a wide range of research applications, for example exploring AI-assisted interventions, analysing qualitative data, developing conversational agents, or testing human–AI interaction scenarios, as demonstrated in this research project: BuddyGPT.

Several models are availeble, such as gemma-4-26b-14b, ministral3-14b-instruct-2512, qwen-3-coder-next (more details are available in the further support section). On request and in cosultaion with BDSi almost any model using or that is adapted to GGUF and llama.cpp can be made available.

How can the Framework PCs be used at the BMS Lab?

The BMS Lab provides several Framework PCs equipped with a range of AI models and tools. Depending on your research needs, there are multiple ways to access and use these systems.

Chat UI Access

Use a browser-based chat interface to interact directly with available AI models. 

Best suited for tasks such as: 

  • General AI experimentation 
  • Exploring model capabilities without technical setup 
  • Ideation and refinement of research proposals 
  • Anything you might use ChatGPT for, but prefer to keep local 
API Access

Access models programmatically through API endpoints. 

Best suited for tasks such as: 

  • Replicable qualitative analyses (e.g., summarizing conversations, topic modelling, classification, etc.)  
  • Creating AI-powered agents (e.g., chat agents (furhat, chatbuddy), ai-assisted web scraping, etc.)  
  • As a local alternative to commercial cloud hosted models for 3rd party tools (e.g., coding agents)  
Borrowing or Dedicated Access to a Framework PC

Researchers can request exclusive remote access to a dedicated Framework PC for implementing and testing more advanced AI workflows and software stacks. Depending on the project and availability, physical access or temporary relocation of the device may also be arranged. 

Best suited for: 

  • Complex applications that require multiple interacting services (e.g., a prototype research application requiring a large language model, database, embedding model, web server)  

Best practices

Accessing and using the framework PCs

To gain access to the framework PCs, please register a request and indicate that you will need to use the framework PCs in the description field. Our team will asses your request and you will receive follow up information. 

Choosing a model

We have three models available that are always loaded, focusing on mainstream and efficiency.

  • gemma-4-26b-a4b: Fast and efficient model from Google. Excellent efficiency with accuracy nearing commercial frontier models, whilst maintaining high speed on local hardware. This is the default model for the chat service.
  • ministral-3-14b-instruct-2512: Europe's answer to US and Chinese models. Trained on a wider range of European languages, trading some accuracy on mainstream tasks in English for better performance in other European languages, including Dutch. The 'instruct' versions of this model are tuned for fast conversation-style responses, and do not support reasoning.
  • qwen-3-coder-next: Popular model tuned for coding and tool use. Developed by Alibaba. Good for long-term reasoning, programming, and agentic use. Popular as a local model powering AI coding tools.

In addition, we have a few modesl available on request and pending approval. Loaded on demand, niche applications and/or large size.

  • mistral-small-4-119b-2603: Larger and more recent version of the Mistral model. Best for complex prompts in non-English European languages, if the ministral models are inadequate.
  • ministral-3-14b-reasoning-2512: Reasoning version of ministral-3-14b. Best for complex prompts or long-running tasks where a chain of thought must be maintained, particularly for non-English European languages.
Asking for help and reporting issues

In case of issues or needed assistance when it comes to use of the framework PCs please get in touch with bdsi@utwente.nl. 

You are responsible for your data.

You must ensure you follow the UT and BMS data guidelines and policies. When using the framework PCs, you are responsible for downloading and safeguarding your own data. 

Make sure you are following the BMS and UT AI use guidelines and policies

Researchers using the framework PCs must comply BMS and UT policy on using AI. 

Close
Suggestions