Frequently asked questions.

vona is a full-service web agency from Leipzig focused on AI integration, software development (web, apps, desktop), design and SEO/GEO. We help companies integrate artificial intelligence sensibly into their existing processes — from simple chatbots to fully automated AI workflows. You can find an overview on the AI integration page.

We work with SMEs, startups and established companies that want to modernize digital processes or use AI technology for the first time. Whether you need a simple website, a company chatbot or a complex AI automation solution — we scale our services to your requirements.

Every project starts with a free initial call in which we understand your requirements and goals. After that we prepare a concrete concept and a transparent quote. Once you approve, development, testing and launch follow — with regular check-ins and interim results. More on About vona.

It depends on scope: a simple AI chatbot is often ready in 2–4 weeks, a website in 4–8 weeks, more complex AI automation or custom software takes 2–6 months. We clarify the timeline in the initial call and record it transparently in the quote.

No. We are based in Leipzig and work remote-first with clients in the German-speaking region and beyond. We hold conversations by video or, where it fits, in person. You can find all ways to reach us on the contact page.

After launch we monitor the solution, apply updates and security fixes and develop it further on request. We agree on maintenance and support transparently so your solution keeps running reliably.

Usage rights and the handover of the source code are set out transparently in the quote — documented and built so that you do not depend on us.

A large language model is an AI system trained on huge amounts of text, which lets it understand and generate natural language. LLMs such as the GPT models (OpenAI), Claude (Anthropic) or Gemini (Google) are the basis of modern AI assistants, chatbots and automation solutions. They learn statistical patterns in language — not “understanding” in the human sense, but impressively powerful pattern recognition.

All three are powerful LLM families from different providers: GPT from OpenAI is broadly applicable and strong at code and multimodal tasks. Claude from Anthropic stands out for long context windows and safe, nuanced writing. Gemini from Google is deeply integrated into the Google ecosystem and strong at research and data analysis. Strengths shift with every new model generation — which model fits best depends on the use case. We help with the choice and are vendor-independent.

Yes. With tools such as Ollama, open-source models (for example Llama, Mistral or Gemma) can be run locally on your own hardware — no data leaves your company. Local models are smaller than the large cloud models but sufficient for many business applications and offer maximum control over your data.

RAG is a technique in which an LLM accesses an external knowledge base during the request. Instead of training all information into the model, relevant documents are searched dynamically and passed to the model as context. That way a chatbot can “know” your current product catalogs, manuals or internal policies — cheaper, more up to date and safer.

As a rule of thumb: when recurring, language-based tasks tie up more than about 5 hours per week — document processing, answering emails, customer support, internal research — and enough data is available. AI does not pay off for every use case, and we will tell you clearly when it does not (yet) make sense.

AI integration means embedding existing models (for example from OpenAI, Anthropic or Google) into your own processes — fast, cost-efficient and immediately available. AI development means training your own models from scratch — only worthwhile for very specific applications and large budgets. For the vast majority of business applications, integration is the right way.

An AI agent is a system that plans and carries out multi-step tasks on its own — for example gathering information, using tools and interfaces and passing on results. In practice we use agents with clear rules, approval steps and logs so that people stay in control. More in the blog article How AI agents change workflows.

An AI chatbot is a conversational interface that holds natural conversations based on an LLM. Areas of use: customer service (24/7 answers), internal knowledge management (employees ask about internal processes), lead qualification (initial consultation and appointment booking), e-commerce (product advice, order status) and more.

Partly. A well-configured AI chatbot can handle a large share of typical support requests autonomously — in one of our projects it is 68%. Standard questions about opening hours, order status or FAQ topics are particularly suitable. Complex, emotional or escalated cases should still be handed to people. The best strategy is a hybrid model: AI for volume, people for complexity.

Via RAG (see above) or fine-tuning. With RAG, your documents (PDFs, Word files, websites, databases) are indexed and searched on every request. With fine-tuning, a base model is trained further on your specific data — more effort, but worthwhile for very specific domains. For most business applications RAG is the faster and more cost-effective solution.

AI automation goes far beyond classic workflow automation: documents are automatically classified, extracted and processed further, emails analyzed and answered, reports created and data from different sources combined. It can help wherever manual work on texts, data or decisions occurs today. You will find examples in our case studies.

Yes. In one of our projects the AI reads the complete email thread and writes a draft reply that the team checks and sends with one click. Every approval and correction flows back as a learning signal; for requests where the system is reliably right, it can be switched to autopilot step by step. The team saves over 70% of the handling time. More in the case study.

RPA (robotic process automation) carries out rule-based, repetitive click and typing tasks — rigid and without understanding of the content. AI automation understands context, can handle unstructured data and makes decisions based on content. Often a combination is ideal: RPA for structured processes, AI where interpretation is needed.

n8n is an open-source workflow automation platform — comparable to Zapier or Make, but self-hostable and much more powerful for complex AI workflows. With n8n we connect LLMs, databases, APIs and email systems into intelligent automation pipelines. When n8n runs on your own infrastructure, all data stays under your control.

OpenAI is the company behind the GPT models. ChatGPT is the end-user interface, the chatbot millions of people use every day. For professional integrations we use the OpenAI API: it lets GPT models be embedded directly into your own applications, chatbots or workflows — controlled, scalable and with your own branding. ChatGPT and API access are two different products from the same provider.

Nano Banana is the name of Google’s AI image model from the Gemini family. It creates and edits images based on text and reference images and suits product images, motif variants and image editing. Nano Banana Pro is the more powerful variant. Among other things we use the model in automated image pipelines for online stores and in our desktop app RENEKI. More in the glossary.

OpenRouter is a unified API through which a very large number of LLMs from different providers (OpenAI, Anthropic, Google, Meta and more) can be reached via a single interface. This enables switching models without code changes, cost optimization by choosing a suitable model and more resilience. A valuable tool for productive AI applications.

Ollama makes it easy to run open-source LLMs locally. We recommend it especially when (1) privacy is critical and no data may go to the cloud, (2) running costs for cloud APIs become too high or (3) an offline solution is needed. Modern open-source models are now very good for many business applications.

GEO is the evolution of classic search engine optimization for the AI age. While SEO aims at good placements on Google, GEO is about appearing in the answers of AI systems such as ChatGPT, Perplexity or Google’s AI Overviews. These systems cite sources differently from classic search engines — with different content strategies and structures. We help you be visible in both worlds. Everything on the SEO & GEO page.

SEO optimizes for classic search engines: the goal is good placement in the results list. GEO makes sure AI systems understand your content, use it as a source and mention it. The two complement each other: good content, clean technology and structured data help in both worlds. More in the blog article GEO: visibility in the AI era.

GEO does not yet have standardized metrics like SEO. Measurable are direct mentions in AI answers (manual monitoring in ChatGPT, Perplexity, Gemini), the share of visits from AI sources in web analytics and visibility in Google AI Overviews. We combine manual monitoring with tracking tools and actively develop this field further.

Usually it comes down to one of four reasons: (1) robots.txt blocks AI crawlers, (2) missing structured data (JSON-LD), (3) too few E-E-A-T signals — authority and trust as a source, (4) the model is based on older training data. For systems with live search such as Perplexity, technical SEO helps directly. For trained-in content, external mentions and links from trustworthy sources are needed.

Yes: native iOS apps, cross-platform apps for iOS and Android and desktop apps for macOS and Windows. As references you will find apps for sportswear, fitness and our own desktop app RENEKI on the app development page.

Native apps make the best use of a platform and feel especially fluid. Cross-platform solutions such as Flutter or Ionic save time and budget when you want to reach iOS and Android at once. We recommend what fits your project.

We choose technology to fit the task: for example TypeScript, React, Vue.js, Node.js, PHP with Laravel, MySQL or PostgreSQL and Docker — and for AI, depending on the task, models from OpenAI, Anthropic, Google or Mistral. You will find an overview with context in the glossaries on AI technologies and web & app technologies.

Yes. Design and development come from one source at our agency: from corporate design through web design and UI/UX to print and print preparation. That way no details get lost between draft and implementation. More on the design page.

It depends on the scope. Decisive factors include functionality and number of channels, the connected data sources and systems (for example knowledge base, CRM or shop), privacy and security requirements and the running costs of the model used. After a free initial call you receive a transparent quote — at a fixed price for clear requirements.

It depends on the scope: a company site is a very different project from a web application or an online store with interfaces. Decisive are the number of pages, functions, design effort, content and integrations. After the initial call we prepare a transparent quote. What matters is not only the price but the return: a well-built website brings inquiries and pays off.

Depending on the project, company size and location there are federal and state funding programs for digitalization and AI. The funding landscape changes regularly: programs expire or are adjusted. A good overview is provided by the federal funding database (in German); chambers of commerce and local business development agencies also advise. We support with the technical description of your project but do not replace funding advice.

Yes, if they are planned properly. Three proven approaches: (1) Local models (for example via Ollama) — no data leaves the company. (2) EU data centers with a data processing agreement. (3) Data minimization and pseudonymization before data is passed to an API. Privacy is a requirement for us, not an afterthought — you should also clarify legal questions with your data protection officer. More in the blog article AI and data protection.

According to the large providers, data sent via the API is usually not used for model training. The exact conditions differ by provider and contract and change — we check them project by project. Sensitive company or customer data should be anonymized before it is passed on; for maximum control over data we recommend local models or EU data centers with a data processing agreement.