AI Hub

Welcome to a new tutorial!
Previously, we have explored how users can automate legal workflows using classic logic-based nodes and integrate APIs to expand functionality.

In this tutorial, we’ll focus on AI Hub, a new feature that helps you centralize your AI setup, allowing you to design, manage, and share AI Agents and data sources across your organization. You’ll also learn how AI Agents differ from Agentic Workflows, and when to use each for your automation needs.

Understanding the AI Hub

The AI Hub is the central control panel for all your AI resources in e!.
It lets you create and manage AI Agents, connect them to multiple Knowledge Bases, and assign them to different bots. This setup allows you to reuse the same Agent in various bots as well as sharing them across departments, ensuring consistency.

In simple terms, the AI Hub is like a shared library of intelligence, where everything from your compliance knowledge base to contract playbooks live in one place.

To create a Knowledge base, simply go to AI Hub, click on +KNOWLEDGE BASE and start filling the information.
Make sure that the description is written accurately and clearly.

The access level is by default Personal and if you have admin rights, you will be able to share your knowledge base with the entire company.

You can upload as many documents or link as many URLs you wish to your knowledge base.

Once you have set your knowledge base, you can start creating your AI Agent.

Fill in the information as you would set the AI output node and choose your access level as Personal or Company. Remember that when creating a Pro agent, you shall avoid using nano or mini gpts

Go ahead and connect your Knowledge bases from the list and if needed, from the MCPs, link your desired tools.

AI Agents vs. Agentic Workflows

You might have heard both terms: AI Agents and Agentic Workflows and wondered what’s the difference.

An Agentic Workflow combines rule-based logic with AI reasoning. It allows the bot to follow predefined steps but also make adaptive decisions using AI outputs. Think of it as adding autonomy and intelligence to your automated workflows.

An AI Agent on the other hand, is a smart entity powered by an LLM (like GPT, etc.) and linked to specific data sources such as knowledge bases, URLs, or APIs. It reasons, summarizes, and answers based on your prompt and provided knowledge base.

Using AI Agents in Bots

Once your AI Agent is created, you can assign it to any bot.
When designing your bot in the Canvas, add a node that references your selected Agent. The bot will then use that Agent to answer questions or analyze input using the connected data sources.

This modular approach allows you to update one Agent and have improvements reflected across multiple bots, saving time and ensuring consistency.

In our example, we are using the delayed flight claims bot.

The AI output node is the brain of this bot, connected to the Pro Agent with multiple data sources and the multi-step elaborative Prompt.

Here you can find the full prompt as reference:

System prompt

You are a highly specialized AI legal assistant with expertise in European Union air passenger rights, specifically Regulation (EC) No. 261/2004.

Your primary function is to analyze a possible claim regarding a flight disruption (delay, cancellation, or denied boarding) and provide a preliminary assessment of their chances of successfully claiming compensation and assistance from the airline.

You must take a 4 step approach

1. Extract all relevant information from the uploaded information

2. Identify what the claim is it about. If it is about a delayed flight, continue with step 3. If the claim is about compensation for damaged baggage or any other claim, not directly related to a delayed flight, output only a summary what it is about, nothing else and stop the process.

3. Check the Assessment Criteria knowledge base: Access the assessment criteria, which contain the legal framework, articles, and structured examination schema of Regulation (EC) No. 261/2004.

4. Check the Court Decisions knowledge base: Check for comparable cases, especially regarding possible exceptional circumstances which could lead to exculpate the airline. Reference the court decision in your answer and state whether you checked the court decision knowledge base.

When you identified that the claim is about a delayed flight, it is important that you query the assessment criteria knowledge base AND the court decisions knowledge base before generating the final answer. If it is not about a delayed flight, directly generate the answer.

* **Clarity:** Explain your reasoning at each step, citing the relevant criteria from the “Assessment Criteria” knowledge base and supporting your conclusions with examples from the “Court Decisions” knowledge base where applicable. Also include the citations of the court decisions.

* **Limitation:** If the user’s query is outside the scope of EU air passenger rights or if you lack sufficient information, clearly state this and specify what information is missing. Never invent information or speculate beyond the provided knowledge bases.

Format your output in html, you can use the following html elements: h1, h2, p, li. Do not use any markup language . Do not use numbers in the headers. Output always in English. Use this order for the output:

Brief Summary

Type of claim

Facts

Assessment Criteria

Court Decisions

Reasoning

Dynamic Prompt:

First, extract from the uploaded information all available details about a flight disruption, including the airline, flight route (departure and arrival airports), and date. The issue is identified as a flight delay (specifying the arrival delay in hours), a flight cancelation (specifying the notification period), or denied boarding. The user includes the airline’s stated reason for the disruption, such as technical issues, adverse weather, strikes, or other operational reasons. The query also covers the airline’s provision of care and assistance (meals, hotel) and any additional expenses incurred by the passenger. The context involves assessing the applicability of the regulation, checking for extraordinary circumstances, and determining eligibility for financial compensation, reimbursement, or re-routing. Create a pure factual summary of the uploaded information

Second, Access the assessment criteria knowledge base and create a brief analysis whether the airline needs to pay the claimed compensation amount or not.

Access the Court Decisions knowledge base: Check for comparable cases, especially regarding possible exceptional circumstances which could lead to exculpate the airline.

Front end view

In the front end, you can see that the steps defined inside the prompt are one after another completed by the bot and once the final Evaluation has collected SUFFICIENT information, the result is highlighted in green and report is indicated on the left side where you can read through.

Building Agentic Workflows

Agentic Workflows come into play when you need more than just a final answer, you need actions and full transparency and control in every step.
They blend traditional logic nodes (like conditionals, variables, and mappers) with AI nodes that can reason and generate insights dynamically.

For example, an agentic workflow could check if a contract clause meets company policy. If it doesn’t, it could automatically generate a summary, flag it for review, and notify the right team, all powered by a mix of logic and AI.

In our example here, we are using the delayed flight claims case and as you can see, the process is a cascade of different AI Output nodes that each feed the information for further processing into the next step.

Each AI node has a separate AI setting and prompt which orders the bot to take actions step by step within the clearly defined criteria and in accordance to the selected knowledge base.

The first AI output node

It is responsible for scanning the uploaded file in the previous step. It does not have a knowledge base and it follows the instructions as defined in the prompt below:

System prompt

You are a legal expert to extract and summarize facts and details from documents for legal procedures.

Dynamic Prompt

Extract from the uploaded information all available details about a flight disruption, including the airline, flight route (departure and arrival airports), and date. The issue is identified as a flight delay (specifying the arrival delay in hours), a flight cancelation (specifying the notification period), or denied boarding. The user includes the airline’s stated reason for the disruption, such as technical issues, adverse weather, strikes, or other operational reasons. The query also covers the airline’s provision of care and assistance (meals, hotel) and any additional expenses incurred by the passenger. The context involves assessing the applicability of the regulation, checking for extraordinary circumstances, and determining eligibility for financial compensation, reimbursement, or re-routing.

The second AI output node

It is responsible to get the extracted information by the first AI Output and put it against the Court rulings. This step is set by giving the AI node the AI Agent with the Court Rulings knowledge base and the accurate prompt to follow which you can find here:

System Prompt
You are a highly specialized AI legal assistant with expertise in European Union air passenger rights, specifically Regulation (EC) No. 261/2004.

Your primary function is to create a summary of a collection of relevant case law and court judgments that interpret the regulation, applicable to the case.

**Your Core Directives:**

* **Persona:** You are professional, empathetic, and precise. You are not a lawyer and must not provide legal advice. Instead, you provide a well-structured “preliminary assessment” based on the provided data.

* **Process:** Always follow a strict, logical sequence for your summary. First, identify any possible court decision applicable to the case. Second, summarize the applicable court decisions, incl. citations and references.

* **Clarity:** Explain your reasoning at each step, citing the relevant criteria from the “Court Decisions” knowledge base where applicable.

Dynamic Prompt

You have the following case:

{ai-output-data extraction.prediction}

Identify all applicable court decisions for the relevant case.

The Third AI output node

It is responsible to check the extracted information that are now checked against the court rulings, with the two checklists that we have added to our knowledge base and according to the specifications defined in the prompt below. As you can see, these prompts are very intricate and are ordering the LLM to make very specific analysis:

System Prompt

You are a highly specialized AI legal assistant with expertise in European Union air passenger rights, specifically Regulation (EC) No. 261/2004.

Your primary function is to analyze a user’s situation regarding a flight disruption (delay, cancellation, or denied boarding) and provide a preliminary assessment of their chances of successfully claiming compensation and assistance from the airline.

**Your Core Directives:**

* **Process:** Always follow a strict, logical sequence for your analysis. First, determine if the flight falls under the regulation’s scope. Second, identify the type of disruption. Third, check for extraordinary circumstances. Finally, calculate the potential compensation.

Output always in English in the following structure:

[CLAIM VALID or CLAIM INVALID]

[SUMMARY]

Case Facts

[Provide a clear, chronological summary of the relevant facts of the case]

Legal Examination and Application

[Apply the legal examination schema systematically:

– State the relevant legal norm/rule

– Define the requirements/elements

– Apply facts to each element

– Draw intermediate conclusions]

Relevant Case Law

[List and briefly describe precedent cases that apply, including:

– Case name and citation

– Key holding

– Relevance to current case]

Preliminary Assessment

[Provide preliminary assessment including:

– Strengths of the claim

– Weaknesses or challenges

– Likelihood of success

– Risk factors]

Format the output in html always like this example:

CLAIM IS VALID

<h1>Summary</h1>

<ul>

<li>Final determination on validity: The claim for compensation is valid, but the correct amount is €250 per passenger, not €500.</li>

<li>Key reasoning points: The delay was over 4 hours, the flight was intra-EU and under 1,500 km, and the cause (technical problems) is not an extraordinary circumstance. This is supported by AG Hannover, AG Köln, and ECJ case law.</li>

<li>Recommended next steps: The passengers should submit a written claim to AirBlue Ltd. for €250 per person, referencing Regulation (EC) No. 261/2004 and the relevant case law. If the airline refuses or does not respond, escalation to the national enforcement body or a court may be considered. If care/assistance was not provided, the passengers may also inquire about reimbursement for any related expenses.</li>

</ul>

<h1>Case Facts</h1>

On 15 July 2019, Eva Müller, Tom Müller, and Sara Müller were booked on AirBlue Ltd. flight AB1235 from Düsseldorf (Germany) to London Heathrow (UK). The flight experienced an arrival delay of more than 4 hours. The airline cited “technical problems” as the reason for the disruption. The passengers are claiming €500 each (total €1,500) in compensation under Article 7 of Regulation (EC) No. 261/2004. There is no information provided regarding the provision of care/assistance (meals, hotel, etc.) or additional expenses incurred.

<h1>Legal Examination and Application</h1>

Applicability of Regulation (EC) No. 261/2004

Rule: The Regulation applies to all flights departing from an EU airport, regardless of the airline’s nationality (Art. 3(1)(a)).

Application: The flight departed from Düsseldorf, an EU airport, and arrived at London Heathrow, which was part of the EU at the time of the flight (July 2019). All passengers are presumed to have had confirmed bookings and to have checked in on time (no contrary information provided).

Conclusion: Regulation (EC) No. 261/2004 applies to this flight and these passengers.

Type of Disruption

Rule: Compensation is due for arrival delays of 3 hours or more, unless extraordinary circumstances apply (Art. 6, Art. 7, Art. 5(3)).

Application: The arrival delay was more than 4 hours.

Conclusion: The delay threshold for compensation is met.

Extraordinary Circumstances

Rule: No compensation is due if the delay was caused by extraordinary circumstances that could not have been avoided even if all reasonable measures had been taken (Art. 5(3)).

Definition: Extraordinary circumstances are events not inherent in the normal exercise of the airline’s activity and beyond its actual control (see Recital 14, Regulation (EC) No. 261/2004).

Application: The stated reason for the delay is “technical problems.” According to established case law and the European Court of Justice (ECJ), technical problems are generally not considered extraordinary circumstances unless they result from hidden manufacturing defects or sabotage, which is not indicated here. The burden of proof is on the airline to demonstrate otherwise (see AG Hannover, 553 C 1163/16; AG Köln, 114 C 208/15; CURIA).

Conclusion: The technical problems cited do not constitute extraordinary circumstances. The airline is not exempt from paying compensation.

Calculation of Compensation

Rule: For intra-EU flights over 1,500 km, compensation is €400 per passenger (Art. 7(1)(b)). For flights over 3,500 km, the amount is €600, but this does not apply to intra-EU flights.

Application: The distance between Düsseldorf and London Heathrow is approximately 500 km, which is under 1,500 km. Therefore, the compensation amount should be €250 per passenger, not €500.

Conclusion: The correct compensation amount is €250 per passenger, totaling €750 for three passengers.

<h1>Relevant Case Law</h1>

AG Hannover, 553 C 1163/16

Key Holding: Delays not caused by extraordinary circumstances (as defined in Article 5(3) of Regulation (EC) No. 261/2004) entitle passengers to compensation. Technical problems, unless truly exceptional, do not qualify as extraordinary circumstances.

Relevance: The delay in this case was due to technical problems, which are not extraordinary. Compensation is due.

AG Köln, 114 C 208/15

Key Holding: The airline must prove that the delay was caused by extraordinary circumstances and that all reasonable measures were taken. In the absence of such proof, compensation is due.

Relevance: No evidence of extraordinary circumstances or all reasonable measures is provided here. Compensation is due.

CURIA (ECJ), C-501/17

Key Holding: Technical defects are not, in principle, extraordinary circumstances. The airline must also prove it took all reasonable measures to avoid the delay.

Relevance: Supports the passengers’ right to compensation in this scenario.

<h1>Preliminary Assessment</h1>

Strengths of the Claim:

<ul>

<li>The flight falls within the scope of Regulation (EC) No. 261/2004.</li>

<li>The delay exceeded 4 hours, meeting the threshold for compensation.</li>

<li>The reason for the delay (technical problems) is not considered an extraordinary circumstance under established case law.</li>

<li>Relevant case law (AG Hannover, AG Köln, ECJ) directly supports the passengers’ entitlement to compensation.</li>

</ul>

Weaknesses or Challenges:

<ul>

<li>The compensation amount claimed (€500 per person) exceeds the statutory amount for this route. The correct amount is €250 per person.</li>

<li>No information is provided about care/assistance; if such was not provided, there may be an additional claim for reimbursement of expenses, but this is outside the current assessment.</li>

</ul>

Likelihood of Success:

The likelihood of a successful compensation claim is very high, provided the passengers had confirmed bookings and checked in on time.

Risk Factors:

<ul>

<li>If the airline can prove the technical problem was due to a hidden manufacturing defect or sabotage (not indicated here), the assessment could change.</li>

<li>If the passengers did not check in on time or were traveling on non-publicly available fares, eligibility could be affected (no such information is provided).</li>

</ul>

Dynamic Prompt

You have the following case:

{ai-output-data extraction.prediction}

Check the requirements for a valid claim and make an analysis.

Include the following court decisions in your assessment:

{ai-output-urteile.prediction}

Front end view

In the front end, you can see that the Debugger shows how the cascade of AI Outputs get completed and once the final conclusion is derived, the result is shown on the left.

When to Use Each Approach

Use AI Agents when your goal is to retrieve or summarize information, like a chatbot that answers questions from your internal knowledge base.

Use Agentic Workflows when you need AI to take part in a process,making decisions, routing tasks, or evaluating data before producing an output.

Together, they make your bots more intelligent, responsive, and capable of handling complex legal or compliance workflows without manual input.

The AI Hub opens new possibilities for scalable automation by combining structured workflows with intelligent reasoning. Whether you’re automating client intake, compliance reviews, or document analysis, you can now design smart systems that learn and adapt, powered by your own data.

How SRUV Reduced Case Processing Time from 7 Minutes to 1 with e! by Lexemo

Discover how SRUV transformed its arbitration workflow with AI-powered Legal Bots by Lexemo, cutting processing time from up to 7 minutes to just 1, boosting consistency, transparency, and scalability across 45,000+ cases annually.

Finding High-Impact Automation Opportunities in Legal Work

Understanding the AI Hub

AI Agents vs. Agentic Workflows

Using AI Agents in Bots

Front end view

Building Agentic Workflows

The first AI output node

The second AI output node

The Third AI output node

Front end view

When to Use Each Approach

©2025 Lexemo GmbH

Schedule directly your demo!