REFERENCE · AGENT STACKS
A guide for the team

Agent stacks, a team that thinks together.

An agent stack is a team of specialist agents that you build — each with its own job, its own knowledge, its own capabilities. What holds them together is built-in intelligence that routes every conversation to the right expert, hands the thread to a different specialist when the topic shifts, and brings in a human when the situation calls for it. The more specialists you add, the smarter the stack becomes.

i
A note on terminology.
An agent stack is the system as a whole — your team of specialists plus the intelligence that holds them together. Specialist Agents are the individual experts you build inside it. Capabilities are the tools each specialist can call. So when this page says “the stack,” it means the full assembly. Customers can name a stack anything — James, Sophie, Alia — and give it any avatar.
The team

Specialist agents

A specialist agent is a focused expert hired for one slice of your customer’s journey. Each one has its own brief, its own knowledge access, and its own toolkit. You decide who’s on the team.

1Examples of specialists you might build — a typical stack mixes 6–10 of these
OT
Order Tracker
Status, shipping updates, delivery estimates.
CX
Cancellation Expert
Handles cancellations, applies retention rules.
RS
Returns Specialist
Return requests, labels, refund policy.
TS
Technical Support
Troubleshooting, walk-throughs, fixes.
BA
Billing Agent
Invoices, payments, plan changes.
OG
Onboarding Guide
First steps, account setup, early questions.
The more specialists you add, the smarter your stack. A single generalist agent is always going to be juggling too many roles. Ten focused experts — each great at one thing — outperforms one good-at-everything agent every time.
What every specialist is configured with ↓
2What makes a specialist
Brief

A focused job description

Each specialist gets its own system prompt — who it is, what it knows, how it should behave. Tight, specific, opinionated. No 5,000-line prompt trying to cover every possible situation.

Knowledge

The right sources, switched on

Each specialist’s access to the Knowledge Base and the Handbook is set independently. A product expert might only need the Knowledge Base. A billing agent that processes refunds also needs the Handbook to know your policy. You decide what each one reads.

Capabilities

Tools that match the job

The order tracker gets the order-lookup capability. The billing agent gets payment access. Specialists hold only the powers they need to do their work — nothing more.

Tuning

Easier to improve

When a response misses, you know exactly which specialist owns it. Fix that one — leave the rest alone. The team gets sharper one specialist at a time.

What it knows

What the stack knows

Your specialists draw on two very different bodies of knowledge — one written for customers, one written for the team. Both shape how the stack answers. Only one is ever quoted out loud.

Public quoted, linked, shared

The Knowledge Base

Customer-facing articles — how-tos, policies in their final phrasing, guides written for the people you serve. When the answer to a customer’s question lives here, the stack draws from it, links to it, and points the customer to the source. The Knowledge Base is the public face of what you know — and it’s how the stack helps a customer narrow a fuzzy problem down to a clear, actionable answer.

Internal followed, never quoted

The Handbook

Your operating procedures — how things actually get done inside your business. Refund thresholds, escalation rules, the wording you’d never put on a help-center page. The stack reads the handbook to know how to handle a situation, then speaks in its own voice. Internal phrasing stays internal.

The handbook is followed, not quoted. It tells the agent how we do things — not what to say to a customer. A specialist that reads “process refunds over $200 only with manager approval” will act on that rule without ever saying the words “according to our internal handbook…” out loud.
What it can do

What the stack can do

Capabilities are the difference between an agent that talks about the problem and an agent that solves it. A capability is a connection to one of your real systems — your e-commerce platform, your CRM, your booking tool. You build them, then assign them to the specialists that need them.

1Examples of capabilities you might build — the framework supports any system you can talk to

Look up an order

Pull a status, shipping date, or carrier tracking link directly from your e-commerce platform.

Cancel an order

Process a cancellation request, apply the right refund rule, and send confirmation — without a human in the loop for routine cases.

Verify identity

Confirm a customer is who they say before any sensitive action — order number, email, postal code, or a one-time code.

Update a subscription

Switch plans, pause billing, change a billing date. Whatever the customer can do in your own portal, the stack can do alongside them.

Track a delivery

Hit the carrier’s tracking endpoint and surface real-time status to the customer in plain language.

Book an appointment

Open a slot, write to your scheduling system, send the confirmation. The specialist makes the booking — not just suggests one.

Without capabilities, your stack can talk. With them, it can act. Capabilities are how an Atender agent stops being a polite explanation engine and starts being a co-worker — one who can actually finish the job.
Built-in intelligence

The intelligence around it all

Behind every conversation, the stack is quietly thinking. It reads the first message and sends it to the specialist most likely to handle it well. As the conversation unfolds, the stack can hand the thread to a different specialist when the topic shifts — and bring in a human when the situation needs one.

1Three things the stack handles for you
Routing

The right expert, first time

When a message arrives, the stack reads it and picks the specialist whose brief best matches what the customer needs. The customer never has to choose a topic from a menu.

Re-route

Hand the thread back when the topic shifts

A customer asking about delivery might end up asking about cancellation. When the conversation drifts beyond a specialist’s expertise, the stack quietly hands the thread to a different specialist mid-conversation — without restarting from scratch.

Handover

Knows when to call a teammate

When the situation needs a human — a frustrated tone, a complicated edge case, anything outside the team’s brief — the AI passes the conversation to a teammate. How that handover happens is up to you.

You build the specialists. The intelligence comes with the system. None of this is something you configure as a separate component. It’s the system’s own judgment — and it gets sharper the more your specialists are doing.
Two ways a handover can happen ↓
2When a human is the right answer
Hot

A live handover

The conversation slides directly to a teammate, with full context already in front of them. The customer keeps talking — only the responder changes.

Cold

A briefed handover

The stack writes up what it’s learned — the customer’s situation, what’s been tried, what’s still unresolved — and queues the case for a teammate to pick up when they’re ready.

At a glance

How a stack comes together

Three things you build, one thing that comes built-in. The shape of every Atender agent stack — at any size, in any industry.

An Atender agent stack
The team Specialists
Examples you might build
Order Tracker Cancellation Expert Returns Specialist Technical Support Billing Agent Onboarding Guide
A typical stack runs 6–10 specialists. Add more as your team grows.
What it knows Knowledge
Public — quoted
Knowledge Base
Internal — followed
Handbook
Public language out, internal language never.
What it can do Capabilities
Examples you might wire up
Order lookup Cancel order Verify identity Update subscription Track delivery Book appointment
Wire as many as your specialists need to actually finish the job.

And quietly around all of it — built into the stack, not a thing you configure — the system’s own intelligence is routing each conversation to the right expert, handing the thread back when the topic shifts, and bringing in a human when one is needed.