AIaaS Introduction

AI for Science. Built for Researchers. Run securely in e-INFRA CZ.Copy Anchor Link

CERIT-SC provides AI services hosted within the national research infrastructure e-INFRA CZ. You can use them through a web browser, connect them to your research software, or integrate them into your coding environment.

The platform provides large language models (LLMs) for working with text and code, models that can process images, and embedding models that represent text numerically for search and document retrieval. Available capabilities depend on the model and the service you use.

Choose how to workCopy Anchor Link

Get accessCopy Anchor Link

AI chat and API access are available to users with an active MetaCentrum account. Masaryk University students and staff can also use the university registration route described in the access instructions.

The university registration route provides access to AI chat and API services. Other platform services may require a MetaCentrum account or additional setup.

1. Register or check your account. Follow the access instructions for your institution.

2. Start with a small task. Open Chat AI, sign in through e-INFRA CZ, and select a model. For example, ask it to explain a short piece of code or revise a paragraph.

3. Connect your tools when needed. Follow the API guide or coding assistant guide to use the service in your own working environment.

Before submitting research data, read the data-handling guidance below.

Choose a modelCopy Anchor Link

Models differ in their capabilities, response time, and resource needs. Start with a model suited to your task and evaluate its output on a small, representative sample.

• Text and code models support tasks such as drafting, classification, information extraction, and programming assistance.
• Multimodal models can also process images, where supported.
• Embedding models convert text into numerical representations for semantic search and retrieval-augmented generation (RAG). RAG supplies retrieved documents as context for a model’s answer.

See the model catalogue for capabilities and the live status page for current availability.

Model changes and reproducibilityCopy Anchor Link

Model versions may be upgraded or retired. A specific version is not guaranteed to remain available throughout a research project.

Maintained model aliases, such as mini or thinker, provide names that can continue to be used when the underlying model changes. An alias does not preserve the model’s behaviour or guarantee identical results.

For work that contributes to a publication or other research output:

• Record the resolved model identifier and available model metadata, rather than only the alias.
• Save prompts, relevant input data, generation settings, dates, and outputs in accordance with your data-management requirements.
• Validate the workflow again after a model change.

The API guide explains how to retrieve model metadata. If your project depends on a specific model version, contact k8s@cerit-sc.cz before starting.

Research data and privacyCopy Anchor Link

Standard hosted models run within e-INFRA CZ. External search tools and separately arranged third-party services can involve sending information outside this infrastructure.

Data handling also depends on how you access the service:

• Chat interface: conversations and uploaded files are stored by the service. Review the retention and deletion information before uploading documents.
• Direct API: the documented API route bypasses the chat interface’s conversation storage. This does not, by itself, establish that a workflow meets your project’s data-protection requirements.

The chat interface is intended for general research assistance and is not certified for protected data. Before using sensitive, personal, or otherwise restricted data, contact k8s@cerit-sc.cz to discuss a suitable setup.

Read the AI data privacy policy for details on external services, storage, logging, and retention.

Use AI outputs in researchCopy Anchor Link

Treat model responses as material to evaluate. Check factual claims and references against original sources, test generated code, and verify calculations and extracted data.

For document-based questions, check that the cited passage supports the answer. Supplying a document does not guarantee a correct response.

Before applying an automated workflow to a large dataset, assess its quality on representative examples and define how errors will be identified and handled.

Share computing resources responsiblyCopy Anchor Link

The platform serves many researchers on shared computing resources. Response times and availability can vary with demand.

• Choose an appropriate model. Start with a smaller or faster model for routine tasks, and use more demanding models when evaluation shows a benefit.
• Limit concurrent requests. Keep batch workloads controlled and respect service rate limits.
• Use stable model selections. Do not automatically switch models in response to short-term utilisation figures on the status page.
• Coordinate independent benchmarks in advance. Contact the support team before running them, as required by the Chat AI guide.
• Plan large workloads. Schedule substantial batch processing outside peak hours where possible, and discuss unusually demanding workloads with the support team.

You can review your API token consumption in the usage dashboard. This dashboard does not include chat-interface usage.

Additional toolsCopy Anchor Link

Once you are familiar with chat or API access, explore tools for more specialised tasks.

Check each guide for setup instructions, access requirements, and limitations.

Support and service updatesCopy Anchor Link

Contact k8s@cerit-sc.cz for access problems, workflow advice, or questions about project requirements.

Join the Matrix community to exchange experience and follow announcements:

• #LLM-General: questions and discussion.
• #LLM-Status: service and model updates.

For current model availability, check the status page.

---

ReferenceCopy Anchor Link

Read more details on our e‑INFRA Blog at https://blog.e-infra.cz/