Access & Authorization
[Warning: Text generated using Generative AI technology - Please assess this text critically before using it]
EP GenAI Hub - Frequently Asked Questions
Access & Authorization
Why can't I log in using my SSO credentials?
To access the EP GenAI Hub, you need a provisioned account. If your account has been created, you should have received an invitation email from ai-ep@europarl.europa.eu.
If you haven't received an invitation and would like to request access, please check if places are available and send your request to ai-ep@europarl.europa.eu.
How can I request access to the EP GenAI Hub?
To request access, send an email to ai-ep@europarl.europa.eu including:
- Your DG, service, and team
- Your intended use cases for the platform
We will review your request based on availability and our prioritization criteria. You will receive either an account creation notification or an explanation if access cannot be granted at this time.
Understanding the Platform
What's the difference between EP GenAI Hub and commercial LLMs like ChatGPT?
EP GenAI Hub is the European Parliament's dedicated platform for using Generative AI tools securely. Unlike commercial LLMs, it offers the privacy level required for EP work.
Key advantages:
- Standard tasks: Run on dedicated cloud servers with enhanced security
- Confidential tasks: Run on-premise within EP infrastructure for maximum privacy
EP GenAI Hub represents the safe alternative for using GenAI within the context of your EP work.
What can I do with EP GenAI Hub?
The platform supports a wide range of tasks:
- Content drafting and refinement (emails, documents, correspondence)
- Summarizing and translating texts
- Extracting information from documents
- Analyzing and simplifying complex texts
- Idea generation and brainstorming support
We continuously improve the application to add features and unlock new use cases. Check the models list regularly to see which models are optimized for specific tasks.
Choosing the Right Model
Which model should I use?
Private Models (On-Premise - Maximum Security)
OpenAI-GPT-OSS-12B
- Best for: Detailed analysis with fewer hallucinations in confidential contexts
- Strengths: Powerful reasoning, instruction following, tool use, and agentic workflows
- Note: OpenAI's first open-weight model release since GPT-2
Mistral-Small-3.2-24B
- Best for: Tasks requiring computational efficiency
- Strengths: Enhanced instruction-following, reduced repetition errors, multimodal input (text and images)
- Context: Up to ~128k tokens
- Use cases: Conversational agents, document understanding, coding, structured outputs
Public Models (Dedicated Cloud - Standard Security)
OpenAI-GPT-4o
- Best for: Versatile, general-purpose AI tasks
- Strengths: Handles multiple input/output modalities
OpenAI-GPT-o1
- Best for: Complex reasoning and sophisticated chat
- Ideal use cases:
- Complex problem solving (math, logic puzzles, philosophical questions requiring Chain of Thought)
- Nuanced creative writing (8 personality presets, less robotic than GPT-4o)
- High-stakes decision support requiring maximum intelligence
OpenAI-GPT-o1-mini
- Best for: Speed and efficiency
- Ideal use cases:
- Simple chatbots and routine customer support
- Rapid summarization (emails, meeting notes, articles)
- Real-time applications requiring low latency (<500ms)
Claude Sonnet 4.5
- Best for: Coding and agentic workflows
- Strengths: Frontier-class model for complex reasoning tasks, substantial improvements in context handling and autonomous task execution
- Ideal for: Long-context tasks, processing massive documents or entire code repositories
Features and Limitations
Can I search the internet through the platform?
No, internet search is not currently available. Information freshness varies by model:
- GPT-4o: October 2023
- Mistral-Small-3.2-24B: October 2023
- Claude Sonnet 4.5: April 2024
- GPT-o1-mini: June 2024
- GPT-o1: October 2024
What's the difference between public and private LLMs, and when should I use each?
Public Models (Dedicated Cloud)
- Run on dedicated cloud servers (Microsoft Azure, Amazon AWS Bedrock)
- More secure than average commercial services, but not 100% private
- Suitable for: Standard, non-confidential EP work
Private Models (On-Premise)
- Run within European Parliament infrastructure
- Fully private and secured
- Suitable for: Confidential work
Can I use ChatGPT, Gemini, or Claude for my EP work?
While EP GenAI Hub should be your default choice, there is no formal prohibition on using commercial services for tasks that don't involve any European Parliament documents (except public ones) or data.
Important: Exercise extreme caution and switch back to GenAI Hub for EP-related work.
File Management
What file types can I upload?
The platform supports a wide range of standard office formats:
- Documents: .doc, .docx, .pdf
- Images: .png, .jpeg
- Data: .csv, .xlsx
What are the file upload limits?
- Maximum size per conversation: 15 MB
- Number of files: No strict limit, but total size cannot exceed 15 MB per conversation
My uploaded PDF is unreadable. What can I do?
If you've uploaded a scanned PDF and the results are poor quality, try enabling the OCR (Optical Character Recognition) feature.
AI Assistants
What are AI Assistants and how can I use them?
AI Assistants are specialized tools similar to:
- CustomGPTs in ChatGPT
- Gems in Google Gemini
- Thematic spaces in the EC tool GPT@JRC
Purpose: Assistants are optimized for specific use cases or ranges of tasks, equipped with specialized instructions and dedicated knowledge bases.
Need more help? Contact ai-ep@europarl.europa.eu
I’m trying to sign in using my SSO but I can’t login, why?
To login to EP GenAI Hub using your European Parliament SSO, you first need a provisioned account. If this is the case you should have received an invitation via email from ai-ep@europarl.europa.eu, If you want to have an access you can check if there are some places available and request one by sending a request to : ai-ep@europarl.europa.eu
How can I start using the EP GenAi Hub?
For that you need to be granted an access, you can send a request, letting us know the DG, service and team you are in, what use you expect to do with it to ai-ep@europarl.europa.eu. We will review your request and based on the number of access available, and our prioritisation system we’ll reply to you hopefuly with a account creation email notification or an explanation on why it is not granted.
LLM Usage
What’s the difference between EP GenAI Hub and standard LLMs like CHatGPT?
EP GenAI Hub is the European Parliament dedicated platform for using Generative AI tools. Commercial LLM’s do not offers the privacy level we need and have to use. EP Gen AI Hub runs either on dedicated cloud servers for standard tasks, and on premise servers for the most critical and confidential tasks.
EP GenAI Hub represent the safe alternative for using GenAI within the context of your EP work.
What can I do with EP GenAI Hub?
Everything you can do with LLM models :
- Content drafting and refinement (emails, documents, correspondence…),
- summarizing and translating texts,
- extract information from documents,
- analyzing and simplifying complex texts
- Idea generation, brainstorm support
- …
We are improving the application constantly to add more features and unlock more use cases. Do not hesitate to check the models’ list and what use cases they are optimized for.
What model should I use (GPT OSS, Claude Sonnet? Mistral…)?
Private
OpenAI-GPT-OSS-12B
Strong for detailed analysis, fewer hallucinations. As it is a private model, the best for analysis in a confidentia tl context.
It is designed for powerful reasoning, instruction following, tool use, and agentic workflows
It is OpenAI first open-weight model release since GPT-2.
Mistral-Small-3.2.24B
It balances strong capability with computational efficiency.
It is offering enhanced instruction-following accuracy and reduced repetition errors compared to prior versions. The model supports multimodal input (text and images) and long contexts (up to ~128 k tokens), making it suitable for tasks such as conversational agents, document understanding, coding, and structured outputs including function/tool calling.
Public
OpenAI-GPT4o
It is designed to handle multiple input and output modalities and serve as a versatile, general-purpose AI for both developers and end users
OpenAI-GPT5.1
Best for complex reasoning and chat
- Complex Problem Solving: Use this for math, logic puzzles, or nuanced philosophical questions where "Chain of Thought" is required.
- Nuanced Creative Writing: It has 8 personality presets and is less robotic than GPT-4o.
- High-Stakes Decision Support: When you need the absolute highest intelligence available, regardless of cost/latency.
OpenAI-GPT5-mini
- Simple Chatbots: Customer support bots that handle routine queries.
- Summarization: Rapidly digesting emails, meeting notes, or articles.
- Real-Time Applications: Voice assistants or apps where low latency (<500ms) is critical.
Claude Sonnet 4.5
Best for coding and agents
frontier-class large language model from Anthropic, positioned as their leading AI for coding, agentic workflows, and complex reasoning tasks. It represents an evolution of the Sonnet model line with substantial improvements in capability, context handling, and autonomous task execution.
Long-Context Tasks: Processing massive documents or entire code repositories with high reliability.
Model | Type | Intelligence | Speed | Best Feature |
GPT-5.1 | Proprietary | ⭐⭐⭐⭐⭐ | ⭐⭐ | Deep "Thinking" Mode & Personality |
Claude Sonnet 4.5 | Proprietary | ⭐⭐⭐⭐½ | ⭐⭐⭐ | Coding & Computer Use (Agents) |
GPT-5-mini | Proprietary | ⭐⭐⭐ | ⭐⭐⭐⭐⭐ | Lowest Cost / Highest Speed |
Mistral-small 3.2 | Open Weight | ⭐⭐⭐½ | ⭐⭐⭐⭐ | Best Enterprise Local Model |
GPT-OSS-12B | Open Weight | ⭐⭐½ | ⭐⭐⭐⭐⭐ | Run on Laptop / OpenAI architecture |
GPT-4o | Proprietary | ⭐⭐⭐⭐ | ⭐⭐⭐⭐ | Reliable Legacy Standard |
Can I connect to the internet to make a search?
At the moment, No : You can’t search the internet, so the limit date for information ‘freshness’ in the GenAI Hub varies from model to model:
- GPT 4o : October 2023
- Mistral-Small-3.2-24B : October 2023
- Claude Sonnet 4.5 : April 2024
- GPT 5-mini : June 2024
- GPT 5.1 : October 2024
What is the difference between public and private LLMs in the GenAI Hub and when should I use one or another?
In the EP GenAI Hub:
- Public models are running on dedicated cloud servers (like Microsoft Azure and Amazon AWS Bedrock), this is more secure than the average commercial cloud services you can find on the web, but not offers 100% privacy guaranties, so this is ok for standard and not confidential EP work.
- Private models are running on premise, within the European Parliament infrastructures and are fully private and secured. Ideal for more confidential work.
Can I use ChatGPT, Gemini or Claude in my daily work at EP?
Ideally it is safer to use GenAI Hub by default, but if you have a task to do, that does not involve any European Parliament documents (except the public ones) or data, there is no formal interdiction for using them. Be extremely cautious, and once done, switch back to GenAI hub to continue the rest of your work.
What file types can I upload?
There is a wide range of supported files, you can use all the standard office work related formats like : .doc, .docx, PDF, PNG, JPEG, .csv, .xslx,
Files, Knowledge and Assistants
What is the maximal number of files I can upload?
There is no real limit as such, but there is a size limit of 15 MB for a single conversation.
What is the maximal size of a file to upload?
There is a size limit of 15 MB for a single conversation.
The pdf I have uploaded is unreadable, what can I do?
If you have a scanned PDF and the LLM isn’t providing you with quality results, you can try to turn on the “OCR” features
What are the AI assistants? What can I use them for?
AI Assistants are the equivalent of CustomGPTs in ChatGPT or Gems in Google Gemini. This is the equivalent of the Thematic spaces on the EC tool GPT@JRC. These are specialised assistant, for a dedicated use cases or range of use cases, having specific instructions, specific knowledge.