When planning and introducing chatbots or voicebots, it is advisable to select a technology partner who has knowledge and experience in the area of Large Language Models (LLMs) and the ChatGPT API.

There are a limited number of providers in German-speaking countries that use the ChatGPT API. Below is a comparison of possible solutions and providers for chatbot or voicebot projects that already use the ChatGPT API (as of June 2023).

It should be noted that the list is not exhaustive and not all possible providers are represented. The listed providers were contacted individually and asked about their possibilities with the ChatGPT API. Other providers were also contacted, but were unable to provide comprehensive answers and are therefore not included in this comparison.

Which chatbot providers were compared?

The following providers were compared in this article and in the supplementary report for download:

• DialogBits
• Enterprisebot
• Paretoblabs
• BubbleChat
• Kore.ai
• PlaynVoice

The ChatGPT API – What is it used for?

When looking at the providers, it is noticeable that the ChatGPT API is used in different ways. All providers use ChatGPT directly to process knowledge sources and answer various user queries.

With DialogBits, Bubble Chat and Kore.ai, the integration of ChatGPT has several functions. It supports the generation of answers to user questions based on extensive knowledge databases such as websites or other data sources. At the same time, it supports the user in the general process of chatbot creation, such as generating additional training records for previously defined intents. With DialogBits, a user request can even be forwarded directly to ChatGPT to generate a response based on the extensive knowledge of the OpenAI chatbot.

Bubble Chat also uses the functionality of ChatGPT for automated testing of the bot.

How is the chatbot trained with the ChatGPT API?

The pure use of ChatGPT can lead to errors or unwanted responses. It is therefore important that technology providers offer options for fine-tuning and prompt engineering.

After consulting the providers, DialogBits, Enterprisebot, paretolabs, Bubble Chat and Kore.ai seem to offer the greatest flexibility in terms of fine-tuning and prompt engineering. With Dialogbits and Bubble Chat, the chat responses can even be individually tailored to the user using prompt engineering.

Paretolabs offer a variety of options for fine-tuning in the form of their mid-layer. With the mid-layer, Paretolabs focuses primarily on industries with sensitive data. More about chatbot with ChatGPT API in banking in this article.

Where does the content for the chatbot with ChatGPT API come from?

Most providers offer the option of using websites, PDFs and other documents as the basis for the bot’s knowledge. The knowledge is usually provided in the form of a vector database. It is also possible for customers to deliver their knowledge directly to the provider in the form of a vector database.

With Dialogbits, Enterprisebot and Bubble Chat, the content of websites can be refreshed regularly. This ensures that the chatbot always receives the latest information from the website. Similar functions can also be set at Paretolabs. You can find out more about using the ChatGPT API in combination with your own website content in the article on my own ChatGPT chatbot experiment.

PlaynVoice only retrieves knowledge from previously uploaded PDFs and other documents (as of June 2023).

How is it ensured that the chatbot does not send any sensitive data to OpenAI?

The providers implement additional process layers or prompts to ensure that not all requests are forwarded to OpenAI.

Enterprisebot and Bubble Chat explicitly emphasize that their solutions can anonymize sensitive information from user requests so that only the anonymous part of the request is sent to OpenAI.

Paretolabs also offers a variety of options for anonymizing data in the form of its mid-layer.

What happens to chatbots with rule-based dialogs?

All technology providers emphasize that they continue to offer their customers rule-based or intent-based approaches in addition to the ChatGPT solution. This is particularly relevant when it comes to fixed processes where a language model such as ChatGPT could be a hindrance.

What are the costs of a chatbot with ChatGPT API?

The costs vary greatly, and not all providers were able to provide transparent prices. In general, there is usually a fixed price for a POC and then license costs depending on usage. The license costs depend in part on the number of tokens or requests to OpenAI and are not communicated transparently by everyone.

Enterprisebot states that they can provide an initial POC for companies starting at CHF 20,000. At Paretolabs, prices for a POC start at CHF 40,000.

What is being done to ensure data minimization?

The use of Large Language Models (LLMs) requires extensive data processing, which is associated with high server costs and high energy consumption. The concept of data minimization aims to counteract this. All the providers presented have implemented mechanisms to ensure that not all requests are sent to OpenAI. Only requests that cannot be answered using other methods are forwarded to OpenAI. The methods used by providers to ensure data economy vary. Prompt engineering also appears to play an important role in the majority of cases. Find out more in the report, which can be downloaded free of charge at the end of this article.

Where is the chatbot with ChatGPT API hosted?

The biggest differences between the providers can be seen in the area of hosting. Not all providers offer the option of on-premise hosting. The majority of providers use Microsoft Azure as their hosting platform, although the location of the Azure Cloud can vary. Other cloud providers mentioned are AWS or Hetzner. More details can be found in the download report at the bottom of this page.

Where can the chatbot with ChatGPT API be used?

If a bot is intended exclusively for use on the telephone, the voice-focused solution from PlaynVoice can be selected. For all other companies, we recommend choosing one of the other providers. These offer chats on websites, WhatsApp, telephone, Instagram, email and other communication channels. Bubble Chat has also named the Swiss provider Bekeeper as an integration option.

What additional functions are available and what is planned?

The providers’ roadmaps are all very promising. Details on the planned features can be found in the detailed report at the end of this article.

What references for chatbots with ChatGPT API already exist?

With the exception of paretolabs, all providers have extensive experience in the chat or voicebot sector and therefore also a large number of reference projects.

In terms of reference projects in the ChatGPT sector, Enterprisebot appears to have the most experience as of June 2023. They were able to name three productive chatbots with ChatGPT API. You can find the exact references in the report, which is available for free download below.

All information, facts and data

This was just a small excerpt from a fairly extensive search to compare chat and voicebot providers that already use the ChatGPT API.

It should be noted that the list is by no means exhaustive. Instead, the focus was placed on precise analysis and direct communication with the individual providers.

The entire report is available via the following button.