In this article, you will learn how to create customized AI assistants that are tailored to the specific needs of your business. Our platform allows you to configure powerful assistants that utilize your internal data and perform recurring tasks efficiently.
How do I configure an assistant?
1. Open the Assistants area
Click Create assistant. You can then configure a new assistant.
Note: By default, the assistant you create is private and available only to you. Upon completion, you will have the option to share it with your team.
2. Enter basic information
Enter a name and a short description: A concise name and a clear description help users to quickly understand the purpose of the assistant.
Upload an avatar: An avatar is the visual face of the AI assistant and makes a decisive contribution to the user experience. It helps users to establish a personal and inviting connection with the assistant and encourages interaction.
Entering a welcome message: This can provide a brief introduction to how users can use the assistant effectively. This can include instructions or examples of questions and commands that the assistant understands. This facilitates interaction between the assistant and the user. A friendly and informative greeting creates a positive first impression and can strengthen the user's trust in the assistant.
Entering conversation starters: An introductory instruction can be used to start a conversation with the assistant. It helps to determine the focus of the interaction and obtain relevant information.
Example: Assistant for communication strategy
3. Select AI model
We know that every assistant is unique and has different requirements - that's why we offer you a selection of different AI models. Whether you are looking for precise answers, creative content or customized solutions for your specific needs - with models such as GPT-4, GPT-4o, Mistral Large, Llama 3 and Cohere Command R-Plus, you have the flexibility to always have the optimal tool at hand.
4. Formulate system instructions
System instructions play a decisive role in the configuration of AI assistants, as they form the basis for the behavior and functionality of the assistants. They define how the assistants should react to certain inputs and what tasks it can perform. Precise system instructions ensure that the assistants works efficiently and delivers the desired results.
In nuwacom you have two ways to create system instructions:
Generate System Instructions: Users can click the "Generate" button to have the AI automatically generate detailed system instructions. This method is particularly useful for users who wish to save time or are not sure how to formulate a command. Based on a short description of the requirements, the AI creates a systematic and precise instruction, which improves the quality and efficiency of the assistant.
Example: AI assistant that makes it easier for new employees to join the company.
The user writes the system instructions themselves: Alternatively, users can create their system instructions manually. This option provides more control and flexibility, as the user can directly incorporate specific requirements and details. This method is ideal for experienced users who know exactly what features and responses they want from their assistant.
With a choice between automatic generation and manual creation, users can choose the method that best suits their needs and level of knowledge to achieve optimal results when configuring their AI assistants.
5. Select response style
Select the answer style you want the assistant to use. Precise always gives the most suitable answer, while Creative can also generate creative and unexpected answers. For a brainstorming assistant, for example, Creative would be very suitable, while an FAQ assistant should always give the right answer, which is why Precise would be better.
If you want the assistant to respond to your individual needs, you can adjust the Temperature and Top-p parameters according to your wishes.
The "temperature" influences the creativity of the AI: a low temperature (e.g. 0.2) ensures precise and predictable answers, while a higher temperature (e.g. 0.8) leads to more creative and variable answers.
The "Top-p" value, also known as "Nucleus Sampling", limits the selection to the most probable words in order to increase the consistency of the answer. A low top-p value (e.g. 0.3) selects only the most likely words, while a higher value (e.g. 0.9) allows for a greater variety.
6. Provide contextual knowledge
Upload relevant documents or select them from the knowledge base to provide the Assistant with specific knowledge. This makes interaction with the assistant particularly useful and tailored to your requirements. Activate additional features such as Web search to extend the functionality of your assistant. These options can be switched on or off as required.
7. Test interaction with the assistant
Test your assistant as often as possible during and after configuration. Use the test environment in the right-hand area of the assistant configurator to ensure that the assistant works as desired.
8. Publish and share assistant
Once you have completed and published your assistant, you can either save it for yourself or share it with your team. Click on the 3-dot icon and select the "Make available to all" option in the menu. This makes the assistant accessible to all authorized users within your Space.
💡Tips
A/B testing: When first setting up the assistant, test different prompt formulations or contextual information to identify the most effective approaches.
Feedback loop: Implement a system to regularly collect feedback from colleagues (or from yourself if using privately) and incorporate it into the refinement of the assistant.
Regularly update contextual knowledge: Review and update your assistant's knowledge base and instructions at regular intervals to ensure timeliness and relevance.