Univerx AI
Agentic AI
AI agents that complete multi-step tasks across connected systems: look up, change, book, trigger a process. Several specialised agents, coordinated by an orchestrator, bound by guardrails and approvals.
Building blocks
Agentic AI in detail.
The building blocks at a glance – all in the same workspace and under the same administration.
Act in several steps
Look up, check, change, book, trigger a process: one task, several steps, one result.
Orchestration
An orchestrator hands sub-tasks to specialised agents and brings the results together.
Tools with permissions
Every action in a business system is approved individually, with role, scope and limits.
Human approval
For defined steps the agent waits for a person to agree before it carries on.
Full action log
Every step with time, tool, data and result, traceable for business units and audit.
Choice of model
Choose language models per task, including your own model, governed in Agent Studio.
#From answering to getting it done
An AI agent understands a request and resolves it in dialogue. Agentic AI goes one step further: it breaks a task into steps, uses tools in connected systems to carry them out and sees the task through.
"Please move my house move to the first of the month" becomes several actions: find the contract, check the date, rebook the meter reading, send the confirmation. The agent works through them in order and reports the result.
#Several agents, one task
Not every agent has to do everything. One agent knows the contracts, one the appointments, one the billing. An orchestrator takes the task, hands sub-tasks to the right agents and brings the results together.
- Understand the task. The orchestrator recognises the goal, the parties involved and missing details.
- Plan. It sets the steps and picks the agents approved for them.
- Execute. The agents look up, change and book in the connected systems.
- Ask for approval. For defined steps, such as a credit note, a person decides.
- Close or hand over. Confirm the result or hand over to an employee with full context.
#Guardrails that run alongside
An agent uses only the tools and data actions explicitly approved for it. For each tool you define which data it may read or write and where the limits are, for example on amounts or deadlines.
| Guardrail | What it does |
|---|---|
| Permissions per tool | Read, write or trigger only within the approved scope |
| Human approval | Defined steps wait for a person to agree |
| Topic limits | The agent stays on its task and declines anything else |
| Disclosure | In customer contact every agent identifies itself as AI |
| Action log | Time, tool, data and result of every step |
These building blocks are designed for the EU AI Act requirements on transparency and human oversight. The principles behind them are described under Responsible AI.
#Built and governed in Agent Studio
Agents, tools, approvals and handover rules are created in Agent Studio. Business teams test them with sample tasks before anything goes live, and every change is approved as a version.
You choose the language model there too, per task. One model for short classification, another for longer texts, or a model your organisation already runs. Agents draw knowledge from Univerx Knowledge, citing the source.
#Where agentic AI helps first
- Cases with a fixed flow across several systems: house moves, address changes, rescheduling with follow-up steps.
- Background work: open a case, file documents, trigger the next process.
- Preparation for people: gather the data so the employee only has to decide.
Univerx AI
AI that resolves requests and backs up people.
AI agents resolve routine requests on the phone and in chat, Univerx Assist supports employees live, Agent Studio governs it all. Plus routing, conversation analytics and knowledge; governance is part of the product.
AI Agents
AI agents understand the request, act in connected systems and hand over to people with context. On the phone as a voicebot, in chat as a text agent, always disclosed as AI.
- AI Act
- DSGVO
Univerx Assist
AI assistance in the conversation: suggested replies, relevant knowledge, next steps and automatic summaries for agents. For team leads, an overview of topics and coaching needs.
- AI Act
- DSGVO
Agent Studio
The workbench for AI agents and prompts: build without code, test with real examples, approve through roles and version every change. Built once, usable in every channel.
- AI Act
- DSGVO
Univerx Knowledge
One knowledge base for employees, AI agents and self-service. With semantic search that understands what is meant, and generated answers that rest on your sources and name them.
- AI Act
- DSGVO
Voicebots
AI agents on the phone: they understand natural speech, clarify the request, act in connected systems and hand over to people with context. Disclosed as AI, available around the clock.
- AI Act
- DSGVO
Predictive Routing
Every contact goes to the person most likely to achieve the best result for the chosen KPI. The model learns from past contacts; you choose the goal and keep the rules.
- AI Act
- DSGVO
Conversation Analytics
Transcripts, topics, sentiment, compliance cues and root causes from calls and messages. So you see why customers get in touch, not just how often. Configurable for co-determination.
- AI Act
- DSGVO
- EU
Natural AI voices with ElevenLabs
Studio-quality voices in many languages and accents, fast enough for live dialogue. For voicebots, announcements, notifications and read-aloud. With ElevenLabs as technology partner, always disclosed as AI.
- AI Act
- DSGVO
- BFSG
Real-time translation
Chat, messaging, email and calls in the customer's language, while agents work in their own. Translated both ways in real time, with a glossary for specialist terms and a transcript in both languages.
- DSGVO
- AI Act
Supervisor copilot
AI summarises what is happening in the queues: topics, emerging issues, sentiment trends, compliance risks. It suggests actions, writes the shift summary and answers questions in natural language. Without individual surveillance.
- AI Act
- DSGVO
Authentication & fraud protection
Verify identity at the start of the contact: by app confirmation, one-time code, knowledge question or, with consent, voice traits. The status holds for IVR, bot and agent, and anomalies are flagged.
- DSGVO
- eIDAS
- AI Act
Frequently asked questions
How does agentic AI differ from an AI agent in chat?
A single agent resolves a request in dialogue. Agentic AI completes a task across several steps and systems, often with several specialised agents coordinated by an orchestrator.
Can an agent do something we have not approved?
No. An agent uses only the tools and data actions approved for it in Agent Studio, with the permissions set there. Nothing else is available to it.
Can we use our own language model?
Yes. You choose per task from the available models or connect your own. Which models and regions suit your deployment is something we clarify with you.
How does this fit the EU AI Act?
Disclosure as AI, human oversight, logging and approvals are designed for the requirements of the EU AI Act. They do not replace a legal assessment of your deployment.
Related
ProductsProcess automation & guided workflows
Build flows and guided forms without code: branching, data actions in connected systems, approvals, timers and error paths. In customer contact, self-service and the back office, with versions, test mode and release.
Use casesBack-office automation
Steer back-office work the way you steer customer contacts: task routing with SLAs, workbins and tracking. Semantic² checks the documents and returns findings in euros with their source. A person always makes the decision.
Use casesSelf-service around the clock
AI agents resolve routine requests on the phone and in chat, at any hour. They are disclosed as AI, act only in approved systems and hand over with context as soon as a person is needed.
Use casesCustomer service
Every request on the right channel, with the right person, with the whole context. Routing, workspace, Univerx Assist and knowledge work together in one flow.
IndustriesInsurance
Claims service with calm and pace: notifications on every channel, checks of invoices and policies with Semantic², agreements signed in the conversation and routine requests resolved in self-service.
IndustriesUtilities & telecoms
High volume, many routine requests: AI agents for meter readings, tariffs and bills, outage communication at peak times, payment reminders with consent checks. Documents for your NIS2 obligations on request.
IndustriesPublic sector
The complete platform for citizen service: every channel in one queue, public service numbers and hotlines with AI agents, appointment booking, video, secure messaging and translation. Plus the digital citizen office for a visit to the authority in one session. Designed for social data protection and accessibility.
Guides2 November 2026 · 11 min readIntroducing AI agents responsibly
Disclosure, handover, permissions, logging, approval and measurement: what should be settled before an AI agent goes live in customer service – step by step, with a checklist for the project team.
GlossaryAgentic AI
Agentic AI receives a goal instead of a single instruction. It breaks the goal into steps, uses approved tools and checks interim results. Its value grows, and so does the need for limits and oversight.
GlossaryAI agent
An AI agent is software that understands a request in natural language, acts in connected systems and hands over to a person when a decision or empathy is needed.
GlossaryEU AI Act
The EU's AI regulation classifies AI systems by risk and attaches duties to each class: prohibitions, strict requirements for high-risk AI, disclosure in dialogue. What that means for AI in customer service.
GlossaryHallucination
A hallucination is an AI answer that sounds convincing but is wrong or invented. It follows from how language models work and can be limited, but never ruled out completely.
See Univerx on your own example.
In a demo we show the platform along your channels and requests, from AI agents to video, workforce scheduling and analytics.
Online, in German or English, with no preparation on your side.