AI agents

Create AI agents: no code, with your knowledge and under your control

An AI agent is a specialised colleague: it knows its role, has your material at hand and – if you allow it – can use tools. This page explains what makes a good agent, how to create one in novendix.AI without programming, and where the limits lie that you should deliberately draw.

No programmingTested before releaseHosted in Germany

What an AI agent does in a company

The difference from a normal chat: an agent does not need to be told what to do every time.

A fixed role

Quote writer, GDPR checker, support replies in your tone of voice – the agent already brings instructions, rules and format.

Your knowledge

Price lists, policies and manuals are attached to the agent. Relevant excerpts flow into the answer without anyone uploading the files again.

Tools

Web search, image generation, canvas for longer texts, connections to apps and your own interfaces – each can be switched on per agent.

A memory

Each agent permanently remembers relevant facts from conversations with you – separate from the general memory.

How to create an AI agent in novendix.AI

Describe the role

Name, description and instructions of up to 50,000 characters. Four blocks work well: role, task, rules, format – plus explicit limits such as “If information is missing, ask”.

Attach knowledge and tools

Store files as knowledge, fix the model or use Smart Select, add web search or canvas, and use fixed input fields instead of a free text box if needed.

Test and release

With evals you create test cases that a second AI scores automatically. Only then do you release the agent to your team – private, for the workspace or public in the store.

You can create and publish your own agents from the Business plan upwards. In the Starter plan you use ready-made agents from the store. All details are on Prices and plans.

Everyday examples

Agents we see again and again in mid-sized companies.

Quote assistantKnows your price list and text modules, fills in the Word template

Tender checkerReads specifications and highlights mandatory criteria

Support repliesAnswers in your company’s tone based on your knowledge articles

Policy checkerChecks texts against your internal AI or data protection policy

Onboarding companionAnswers new employees’ questions from the handbook

Form agentFill in fixed fields, submit, done – for anyone who does not want to write prompts

How sales and bid management benefit is described in Proposals and tenders with AI. The difference between a chat and an agent is explained in Chat or agent?

When agents act: set limits deliberately

The more an agent may do on its own, the more important clear rules become.

Own interfaces only with IT. An agent can call your own HTTP endpoints. Every endpoint the AI may use is a conscious decision.
Check write actions. Treat appointments, messages or records via connected apps like a signature: look at them once before they go out.
Mind the visibility. Before an agent goes public, check whether its instructions or knowledge contain internal information.
Automate recurring work. If an agent should work at fixed times or be part of a multi-step process, connect it to tasks and workflows – see AI workflows.
Data stays within your framework. Platform in a German data centre, models in the EU, only models with Zero Data Retention. Details on GDPR-compliant AI.

Agent, task or workflow – which for what?

Three tools that are often confused. The rule of thumb: an agent is a role, a task is an appointment, a workflow is a process.

	AI agent	Task	Workflow
What it is	Specialised assistant with a role, knowledge and tools	An instruction to the AI at fixed times	Multi-step process with a trigger, AI steps and outputs
Who starts it	A person in the chat	The schedule	A person, a schedule or a webhook
Typical for	Quotes, reviews, support replies	Weekly summary, industry news	Sorting incoming items, enquiry → draft quote
Can be combined	As context in tasks	With an agent or project	With AI steps and app connections

How tasks and workflows work in detail is explained on AI workflows.

Good instructions for AI agents

The quality of an agent depends to a large extent on its instructions. Four blocks have proven themselves:

Role. Who is the agent? “You are the quote assistant of company X and write for technical buyers.”
Task. What does it do? “You create a draft quote from the enquiry based on the attached price list.”
Rules. What always applies, what never? “Do not invent prices. If information is missing, ask.”
Format. What does the result look like? “Reply as a Word document using our template.”

Write in full sentences and state limits explicitly. How teams maintain such templates together is described in A prompt library for your company.

Frequently asked questions about AI agents

What is the difference between an AI agent and a chatbot?

A chatbot answers questions. An agent has a fixed role, its own knowledge and can use tools – for example research on the web, create documents or retrieve data from connected systems. More on this in Chat or agent?

Do I need programming skills to create an AI agent?

No. Agents are created in novendix.AI via a form: instructions, knowledge, model, tools. You only need your IT for your own interfaces to specialist systems.

Which models can an AI agent use?

Any model included in your plan – fixed or via Smart Select. An overview is given in AI models at a glance.

Is the agent trained on my documents?

No. The documents are searched when needed and relevant excerpts are included in the answer. Only models that contractually do not train on customer data and do not store requests are enabled.

How do I test whether an agent works reliably?

With evals: you create test cases and criteria, and a second AI scores the answers automatically. Useful before release and after every major change.

What does it cost to create AI agents?

Your own agents are available from the Business plan at €35 per user per month plus VAT, from one user. You can test free for ten days.