Agent Builder

​

What is the Agent Builder?

The Agent Builder is where nuwacom agents are created and configured without code. An agent handles complex, recurring tasks from the initial request to a reviewed result using shared company context.Admins retain central control over permissions, approvals, and logging. Once an agent is saved, it can be called directly in the chat by mentioning it with ”@”.

​

Before you start

Defining how the agent should help produces a clearer structure from the outset. Four questions guide the setup:

• Goal: What should the agent achieve for the user?
• Steps: Which logical sub-steps make up the task?
• Knowledge: Does the agent draw on a single file, multiple documents, or examples?
• Input type: Do user inputs arrive as free text, structured forms, or targeted follow-up questions?

​

Creating an agent with the builder

A new agent is created via the Create Agent option in the Agents module. Three paths are available.

​

Start with a template

nuwacom provides predefined agents: a searchable collection of agent templates that cover proven use cases out of the box. Instead of starting from zero, a matching template is selected and then adapted to the brand and requirements. Templates are organized by business area so the right one is found quickly:

• Sales: for example, Proposal Assistant, Competitive Intelligence Agent
• Customer Support: for example, Knowledge Base Assistant, Customer Feedback Insights Agent, Support Triage Agent
• Marketing: for example, Brand Review Agent, Content Repurposing Agent, Campaign Briefing Agent
• Legal: for example, Data Privacy Review Agent, Contract Review Agent
• Product: for example, Document Intelligence Agent, Product Research Agent, Release Notes Agent

​

Create with AI

Describing what the agent should do in the input field starts an AI-assisted build, and the agent is assembled automatically. This method saves time and helps with formulating the right instructions. Every detail can be fine-tuned manually afterward.

​

Create from scratch

The from-scratch option configures the agent step by step, with name, instructions, knowledge, and behavior defined manually. It is available via the three-dot menu and offers maximum control with specific details from the start.

​

Refining and testing

An agent is tested during creation and after configuration. The “Run” preview shows how the agent works and supports optimizing the instructions. Several iterations are normal. If something does not work as intended, a clearer description or a revision of the relevant section usually helps, and running typical use cases confirms that the instructions hold in different situations.

​

Saving, sharing, and publishing

After configuration, an agent stays private or can be shared with users, user groups, or the whole workspace. When sharing, choose whether recipients have read or edit rights. The share button makes the agent available in the workspace or shares it with specific people.

Important*: Knowledge connected to the agent is shared with those users as well. When an agent with attached prompts is shared, the prompts are shared too. When an agent with actions is shared, permissions also matter: users either set up their own accounts or use the existing account.*

Tip*: After saving, an agent can be pinned to the navigation bar for quick access.*

​

FAQ