Creating and configuring agents

The agent is the one doing the talking. A solution can have several; with no active agent, the chat answers nothing.

The agents list

Open the solution and click Agents.

The list can be shown as a grid or as a tree. The tree view shows the main agent, its specialists and the agents not linked yet — the best way to understand a solution you did not build yourself.

Creating the agent

Click + New Agent and fill in the form.

Field	What to enter
Name	How the agent is identified across the screens.
Description	What this agent is good at. When a solution has several agents, this description is how MarIA decides which one answers.
Instructions	The agent’s behaviour: language, tone, what it should do and what it should refuse. This is the most important field on the screen.
Default Provider	The AI provider the agent uses.
Default Model	The model within that provider.
Active	Inactive agents never answer.

Writing good instructions

Be explicit. State the language of the answer, the tone, the audience, what the agent may assert and, above all, what it must refuse or hand over. A vague instruction produces a generic answer.

Changing the provider

The provider cannot be changed while the agent has files attached. Remove the files first, then upload them again afterwards.

The agent tabs

Once saved, the agent opens with five tabs.

Files

Reference documents the agent consults when answering. You upload the files and follow the status of each one. The accepted types and sizes come from the solution settings, per provider — if an upload is refused, that is the rule at work.

Specialists

Focused sub-assistants that the main agent consults when a question needs depth. Each specialist has a name, description, instructions and can have files of its own. The description is what guides the main agent to consult it at the right moment.

Tools

What the agent can do beyond talking.

Type	What it does
External API	Calls an endpoint in another system and uses the response
Sub Agent	Delegates part of the work to another agent
Cards	Shows formatted content inside (or outside) the chat window
Forms	Collects structured data during the conversation
Map	Displays and lets people interact with points on a map
Web search	Looks up current information, with the sources cited
Image generation	Creates images from the conversation

Every tool has a description for the AI. Write it as an instruction about when to use that tool — it is what determines whether the agent reaches for it at the right time.

MCP

Connects the agent to the external tool servers registered in the solution and lets you pick which of those tools it may use.

Memory

Defines what the agent keeps between conversations, and with which permission — read only, or write as well.

Several agents in one solution

When a solution has more than one active agent, MarIA picks which one answers each message based on the descriptions. So keep the descriptions precise and non-overlapping: two agents described alike compete for the same questions and the result becomes unpredictable.

Once configured, open the chat and test it.