Skip to content
Catalisa / AI Agents

Use AI agents to serve customers and execute tasks.

Define each agent's role and connect information and tools. Automate customer service, system queries and tasks, with team approval when needed.

Explore an agent's roles ↓
CONTEXTTOOLSONE ROLE. PURPOSEFUL CONNECTIONS.
Value emerges when intelligence meets operations.
Intelligence is part of a composition

A conversation can
find a piece of information.
And a next step, too.

An agent uses the available context to interpret a request and decide how to respond. When it has the necessary tools and access, it can take part in the actions that move the journey forward.

Design starts with the work: who it supports, what information it can access and how far it can go. Its actions make sense within that responsibility.

Three questions that give it shape

An agent needs
know its place.

01

The purpose

Goal, tone, instructions and handoff criteria guide how it works.

02

The context

The conversation and authorized information help understand the request.

03

The tools

Defined actions and access connect the agent to what the operation allows.

The role changes with the business

Support during the conversation.
Continuity in the operation.

Explore the contexts and access level.
The same agent concept meets different kinds of work.

Customer service / One possible role

Service that finds the next step.

Understand the request, look up the permitted context and help the person move forward.

Explore the access level
Customer

I need to reschedule my appointment. Is there any time available this week?

Agent / With authorized context

I can check availability so you can choose a new time.

Illustrative conversations. No model or system is queried.
What supports the answer
View-only access

Prepare the rescheduling request

In this demo mode, the agent looks things up and provides guidance. Preparing an action requires the corresponding access.

Context that follows the conversation

Continuing begins
not yet recognized
what has already been said.

Continuity can reuse the conversation history within the configured period. Information from one message helps interpret the next.

This context has limits: session length, available data, and the person's access. The design needs to define when to resume, when to ask, and when to hand off.

ONE CONVERSATION, WITH CONTINUITY
First contact

“I'd like to book an appointment.”

In the same conversation

“I'd prefer the afternoon.”

Context connects the messages.Scheduling
+ preferred time
Illustrative example. Session continuity and duration depend on configuration.
Tools connect intent and capability

The agent finds
the pieces of the work.

A tool provides access to an operation.
Your use must respect your role and permissions.

Autonomy within boundaries

Instructions guide.
Permissions set the boundaries.

Text guidance helps define behavior. Access to actions needs its own controls, configured in the platform and in the journey.

Explore governance in depth
01

Who's asking?

Role and access context determine what can be viewed or done.

02

Which action is available?

Select tools and configure the applicable controls. Naming an action should not mean unrestricted access.

03

What requires a person?

Approvals and sensitive actions need a review experience defined by the operations team.

04

What to do when it doesn't know?

Missing information, uncertainty and failures should be routed onward instead of met with a made-up answer.

The role gets configured

From intent to identity.
From identity to tools.

A view of agent management
and the start of its setup.

Configuration / Agent identityEnlarge ↗

Reference for the agent creation interface, with identity, skills, security, instructions and review steps.

Interface baseline from 07/30/2026 · demo data

Screenshots preserved in the repository. They are interface references, not evidence of performance or of a running conversation. Click to enlarge.

Each part has a role

The agent interprets.
The flow connects.
The product provides context.

Building a complete experience takes more than configuring a conversation. Capabilities and business rules need to come together in the same design.

To shape the engagement

Autonomy starts
with clarity.

Do I need to grant access to every tool?

No. Choose the tools the agent's role requires and configure the applicable access and controls. A well-defined responsibility makes it easier to observe and validate behavior.

Does the agent remember every conversation forever?

No. Continuity depends on the session configuration and the allowed inactivity period. Conversation history also does not amount to unrestricted access to company data.

Is a copilot the same as automated support?

These are different roles. A copilot can support the professional during a conversation and prepare records for review. Automated service interacts directly with the customer, within the defined rules and permissions.

Does the agent decide the business rules?

Rules need to be explicit and live in the responsible capabilities. The agent can interpret the request and use tools, but the design must preserve the operation's policies, authorizations and reviews.

Can I use the systems I already have?

Tools and integrations can connect the agent to information and actions in existing systems. This depends on the available interfaces and the required credentials, contracts and permissions.

How do I start evaluating?

Choose one role and a short journey. Define the goal, permitted data, tools, and handoff points. Validate the behavior before expanding its scope.

From intelligence to the work that matters

Choose a purpose.
Find the connections.

/ Interface reference

0 / 0