Avatar, name, description and category
An avatar gives the agent a visual identity. Name and short description are entered, unless the AI has already done this. An appropriate category ensures clear organization of the agents.Instructions
Instructions define how the agent responds and which tasks it performs. Clear instructions enable effective work and the desired results. Two methods are available: Generate instruction with AIUsing “Create” the AI generates detailed instructions. This saves time and helps with uncertain wording: based on a short description, clear, effective instructions are produced for reliable operation. Write instructions manually
The instructions can also be defined manually. This option offers full control and flexibility to directly include specific requirements and details. Ideal for experienced users who know exactly how their agent should act. The choice between automatic generation and manual creation allows the method that best fits needs and level of knowledge, for optimal results in configuration.
Structure of clear instructions
Clear instructions arise from three key elements: Task, Action and Goal. Supplementary “Response Guidelines” control tone and writing style. Together they define what the agent does, how it responds and how the results are structured. TaskThe role describes who or what the agent should be. This influences tone, expertise and communication style. Example: Act as an automated messaging assistant for Slack that enables users to send messages to specific channels or people. Action
The action describes the individual steps of the agent. This becomes especially important when multiple actions are involved. The clearer the steps, the better the result. Example, if users want to send a message:
- Ask for the channel or recipient.
- Display a preview of the message.
- Send the message to the specified recipient.
- Confirm the sending.
The response guidelines specify how the agent replies, for example regarding tone, writing style or approach. Example:
- Use a friendly and concise language.
- Ensure clarity when confirming recipient and message content.
- Handle errors confidently and offer steps to resolve them if sending fails.
- Keep a log of sent messages upon request.
The goal defines in one to two clear sentences what the agent should achieve for the user. Example: Optimize communication in Slack by providing a fast and reliable way to send messages, in order to increase team collaboration and productivity.
Agent start options
When creating, it is determined what happens when the agent is selected. This guides users from the start. Two options are available: Start actionAutomatically triggers a prompt when the agent is selected, also presented as a form to control user inputs. Ideal for structured, recurring processes. Use ”+ Add prompt” to define the start form. Conversation starters and welcome message
Instead of a fixed start action, optional conversation elements make the interaction more flexible and inviting:
- Conversation starters: Up to 4 different conversation starters appear randomly. They inspire users and show possible questions or actions. Use ”+ Add prompt” to define start prompts and a welcome message.
- Welcome message: A short welcome message invites interaction. It appears as soon as the agent is selected but no action has yet started. Suitable for open, informal interactions without structured input.
Actions
Adding actions from different tools
Actions bind tools like Slack, Google Drive, Microsoft Teams and others. This way agents not only respond but perform tasks: send messages, manage calendars or retrieve files. Complete workflows run directly in the chat.Supported tools include: · Slack: send messages, notify channels · Microsoft Teams: send messages, notify channels · Gmail: draft or send emails · Google Calendar: retrieve, add or update events · Google Drive: retrieve files, add documents · SharePoint: retrieve files, add files · OneDrive: manage documents Multiple connectors can be combined within one agent. The full article shows all integrations and use cases.
Adding nuwacom Actions
In addition to actions from connected tools, nuwacom-native actions can be added — for example, to enable web search or access to the Knowledge Base. To add nuwacom actions:- Open the agent and navigate to Actions.
- Select Add action.
- Choose Nuwacom.
- Add individual actions as needed, or select Add Nuwacom to add all available nuwacom actions at once.
- Search Knowledge Base: Searches the Knowledge Base for relevant information.
- Search Web: Searches the web for current information and answers.
- Write content: Creates and edits content in editor.
- Create email: Creates and edits emails.
- Create Slides: Creates and edits presentations.
- Generate images: Creates and edits images using AI.
- Generate videos: Creates short videos based on text descriptions.
Knowledge
Knowledge enables more accurate, relevant and contextual responses, especially for company-specific information. Example: A support agent accesses stored product documentation and FAQ documents for each response and formulates customer replies solely based on these linked contents. Knowledge comes from the following sources:- Upload files: Upload new documents directly.
- Use “My files”: Select already existing personal files.
- Include knowledge folders: Add entire knowledge folders, individual subfolders or specific files.
Skills
Skills can be attached directly to an Agent to define how it responds. Once a Skill is added, it is automatically active for everyone using that Agent. Users don’t need to install or select the Skill themselves.This means you can provide a fully configured Agent with tone, structure, and rules already built in. Users simply open the Agent and start working, while every output follows the same standard. ExampleAn Agent with a Meeting Notes Skill always structures outputs into Key Decisions, Open Questions, and Next Steps — no setup required for the user. Skills vs. Agents — The Critical Distinction
Combining a Skill with an Agent enables autonomous task execution with consistent output quality. If your team has to navigate somewhere deliberately for something that should just always work, that behavior belongs in a Skill.
Settings
The “Settings” tab adjusts the agent’s behavior and interaction and controls tone, intelligence and availability. Suggest replies automaticallyThe AI suggests a reply after each agent message, promoting the flow of conversation. AI model
A specific model ensures consistent capabilities or responses.
Learn more about AI models here {Link}. Voice
A voice that matches the brand or preferred tone can be selected. Response style
Determines how the agent replies: · Precise: for accurate, fact-based answers · Creative: for brainstorming or open tasks · Balanced: a mix of both
Publish agent externally
An agent can be embedded outside the workspace, for example on a website via widget or iFrame.- Enable external publishing: In the “Publish agent externally” area, turn the switch on.
- Set allowed domains: Enter one or more domains into the whitelist to control embedding locations. Example: https://yourcompany.com. Click “Save” to apply the domain restrictions.
- Choose embedding option: Use iFrame or widget.
Creates a fullscreen conversation experience, ideal for intensive interactions or support. After publishing, the iFrame code is copied and inserted into the website’s HTML. Option B: Widget
Adds a floating AI chatbot that provides quick help without disrupting the user experience. After publishing, the widget code is copied and placed before the page’s closing tag. The embedding code is only available after the agent is published. Test integration
Check on the website whether the agent loads correctly. Verify styling, expected response behavior and effective domain restrictions. If you have questions about embedding, the support team will be happy to help.