Connect your AI assistant

You can connect your own AI assistant — like Claude, ChatGPT, or another one you already use — to your store on Infynia. Once connected, just ask, in plain language: “how’s my cart recovery going?”, “is my WhatsApp number healthy?”, “which products am I losing?”. Your assistant reads your store’s real data and answers.

Before you start

You need the store Owner or Administrator role to connect an AI assistant.
Use an AI assistant that already supports connectors or custom integrations (for example, Claude, ChatGPT, or Claude Code). In ChatGPT, you need a plan that allows custom connectors.

The AI Connector page

Inside Infynia, open Integrations and, on the AI Connector card, click See how to connect (or Manage, if you already have connected apps). The page shows:

the Connector URL, with a Copy button;
step-by-step instructions for each app — Claude, ChatGPT, and Claude Code (for Claude Code, a ready-made command to copy and paste into the terminal);
the list of Connected apps for the store (see Manage your connections).

How to connect

In your AI assistant, open the connectors or integrations settings and choose to add a new connection (in Claude, Add custom connector). If you like, name it Infynia.
Paste the connector URL when it asks for one. It’s shown on the AI Connector page:
```
https://infynia.app/mcp
```
Your assistant will open an Infynia sign-in screen. Sign in with your account.
The Connect Infynia screen shows which app is asking for access and what the connection can read. Only continue if you started the connection from that app. Under Connect which store?, choose the store. If you like, also tick the optional box that lets the assistant read customers’ replies to your broadcasts (see Customer replies to broadcasts).
Click Connect.

That’s it — from there, you can chat with your assistant about that store’s data.

What you can ask

Some examples of what your assistant can answer using your store’s data. You don’t need to use these exact phrases — ask in your own words, as you would with anyone on your team.

Recovery and store health

For e-commerce stores. Stores in CRM mode have no cart recovery: see Sales pipeline (CRM stores).

“How’s my cart recovery going?” — recovered revenue, carts recovered, recovery and response rates.
“Where am I losing people?” — the path of your carts and payments (received, contacted, replied and recovered) by recovery type, and how each follow-up message performs.
“Which recovery template works best?” — how each message template performs and, if you have an A/B test running, which version is ahead.
“What should I do now?” — suggested next steps to improve recovery (for example, an automation to turn on).
“Is anything broken? Why isn’t it sending?” — your store’s problems in order of severity, with an explanation and what to do. This includes WhatsApp number quality problems and send errors explained in plain language.
“Is everything working?” — the status of your connections and whether there were any recent delivery problems.
“Which sources recover the most?” — performance by traffic source (Google, Facebook, Instagram, direct and others).
“How has revenue changed over time?” — recovered revenue and purchases per day, week or month and, if your plan includes broadcasts, the revenue credited to broadcasts. The two numbers can overlap (the same purchase may appear in both), so they shouldn’t be added together.
“How much could I recover?” — while recovery hasn’t been turned on yet, the abandoned carts and open value of the last 60 days and an estimate (not a promise) of what can be recovered.
“Which carts are queued to be sent?” and “What happened to this cart?” — the list from the Dashboard’s Checkouts tab (Dashboard), with each cart’s stage, its next send and why a cart wasn’t contacted, and one cart’s timeline (message scheduled, sent or not delivered with the reason, reply, payment). No customer data: to see who it is, open the cart in the app.

WhatsApp number health

“Is my WhatsApp number healthy?” — the number’s quality and how it has moved over the last 30 days, the connection status, any active restrictions or warnings, the daily limit of new people you can message and how much of it you’ve used.
“Why aren’t my messages arriving?” — the delivery rate over the last 30 days, the most common errors with what each one means and what to do, and delivery by message template.

Customer service

“How is my customer service doing?” — how many conversations the AI resolved on its own and how many were passed to your team, first-response time, conversations waiting, each attendant’s performance and satisfaction survey results (AI and team), compared with the previous period.
“How are my AI agents doing?” — which agents are active and, for each one, conversations handled, how many the AI resolved alone, how many were passed to the team and customer satisfaction.
“How can I improve my agent’s instructions?” — the assistant reads the configuration and instructions (prompt) of your AI agents, cross-checks them with performance and with the reasons conversations were passed to your team, and suggests concrete changes, field by field. You apply the changes yourself, in Infynia’s AI Agent menu.
“What improvements has Infynia suggested for my agent?” — the pending, accepted and rejected agent improvements (on more complete plans).
“How would my agent answer this message?” — the assistant tests the agent with a sample message, like the test on the AI Agent page. Nothing is saved or sent to a customer. This test uses limited AI capacity, so use it sparingly.
“How’s my inbox right now?” — how many contacts are unassigned, need a human, are waiting or unread, and the most used tags. These are live counts, with no conversation text.
“Is my human handoff set up correctly?” — whether the AI hands off conversations, who receives them and how they’re distributed, the handoff topics and the satisfaction survey questions, with warnings about settings that don’t work (for example, nobody to receive).
“What did my team flag as wrong in the AI’s replies?” — how many conversations your team flagged in the period and which reasons were used.

About conversations, the assistant only sees aggregate numbers: never their content or customers’ identities. Service metrics are available on more complete plans.

Sales pipeline (CRM stores)

If your store uses CRM mode, the assistant answers with your sales pipeline data:

“How’s my sales pipeline?” — the value of the open pipeline, win rate, deals won and their revenue, average sales cycle and new deals.
“Why am I losing deals?” — the lost reasons and how many deals fall under each.
“Which deals are stalled?” — how long deals stay in each stage and which ones are aging without moving forward.
“Which deals should I tackle first?” — the most valuable open deals, the ones sitting longest in their stage and overdue follow-ups. They’re shown by deal number only, with no contact name or phone: open the deal in the app to see who it is.

Products and sales

For e-commerce stores.

“Which products am I losing?” — the most abandoned products that were not recovered, and where recovery is worth focusing.
“What are my best sellers, and what’s trending up or down?” — the products that sell the most, trends, how concentrated your revenue is, and suggested actions with their estimated impact.
“What does my customer base look like?” — how many customers you have, how many you can reach on WhatsApp (and how many opted out), how the base has grown month by month, and how many have bought or never bought.
“What customer groups do I have?” — the RFM matrix (recency, frequency and value) with groups such as champions, loyal and lost.
“Is my catalog in good shape?” — the catalog sync status, products with no image or price, out-of-stock products and which of them are still being abandoned (to restock or hide).

Product questions are available on the Basic and Growth plans.

“Which ad or campaign should I scale or review?” — the revenue and sales of each ad or campaign, the trend against the previous period, a “scale”, “attention” or “review” signal with its reason, and the quality of your link tracking.

Configuration

“Which automations are on?” — your store’s automatic messages (recovery, order notifications, follow-ups), whether each one really sends or is stuck (no template, awaiting approval, broken…), which template it uses and how each performed in the period.
“What are my message templates, and why was one rejected?” — the template library with each one’s status and text and, for rejected ones, the reason explained in plain language, with help rewriting.

Plan and costs

“What’s my plan and what’s included?” — your current plan, billing status, renewal and the features included.
“How much of my limits have I used?” — usage of your plan’s limits.
“How much am I spending compared with what I recover?” — the cost of operation against recovered revenue, cost per recovered sale and return, plus an estimate of the current cycle’s bill.
“What were my latest invoices?” — amount, status and period.

The assistant doesn’t show payment methods and can’t change your plan. Exact amounts are in the subscription area of the app.

Broadcasts

“How did my broadcasts do?” — sends, conversions, revenue, cost and return on investment, over the period you choose.
“Which were my best broadcasts?” — broadcasts ranked by reply rate, conversion, revenue or return, each with the audience it used.
“How are my broadcast sequences doing?” — each sequence, step by step: sends, replies, conversions, revenue, how many people drop off between steps and when the next one is due.
“Which broadcast template converts best?” — how each template used in your broadcasts performed, with its text, and help writing the next one within what Infynia’s template creator accepts.
“What’s the best day and time to send?” — replies, conversions and revenue by day of the week and by time, with a warning when there are too few sends to draw conclusions.
“Which audience filters can I use?” — the filters available in your store (segments, abandonment reasons, contact source, forms, products and others) and how many people each one reaches.
“How many people would this broadcast reach, and what would it cost?” — a preview of the audience and a cost estimate, without sending anything.

Customer replies to broadcasts (optional)

By default, your assistant does not see what customers replied to your broadcasts. If you want it to read those replies — for example, to learn what to say or avoid in the next message — you can allow it when connecting:

It’s off by default. If you don’t tick anything, replies stay private.
To turn it on: connect again and, on the store selection screen, tick the box Let the AI assistant read how customers replied to my mass dispatches. The connection screen shows in your browser’s language (Portuguese, English or Spanish).
It applies to that connection only. The permission belongs to the connection you just authorized: another assistant or another app cannot read replies unless you tick the box when connecting them too. In the Connected apps list, connections with this permission show a Reads replies badge.
To turn it off: connect again without ticking the box. Connecting again to the same store replaces that app’s previous connection to that store: the old one stops working immediately. Connections to other stores are not affected.
What it sees: only the text of each customer’s first reply to a broadcast, without saying who replied. Phone numbers, e-mails, documents (CPF/CNPJ) are removed, and links are removed on a best-effort basis (not every link is detected), but names and other personal details customers write may still appear. Shared contact cards show up only as “contact”, with no content. It doesn’t see the rest of the conversation.

Infynia help

“How does [Infynia feature] work?” or “what does [a metric] mean?” — answers based on Infynia’s help guides.

Ready-made prompts

Besides your own questions, Infynia offers ready-made prompts that walk the assistant through a task step by step. Each store only sees the prompts that make sense for its type and plan:

Plan my next dispatch — picks the audience, suggests the message, the day and time, and estimates the cost, based on what has already worked in your store. You can optionally state the goal of the broadcast (for example, “win back customers who haven’t bought in 90 days”).
Weekly store review — a check-up of the week: what’s broken or at risk, WhatsApp health, results (or the pipeline, for CRM stores), how customer service is going and the three most valuable actions to take.
Why am I not recovering carts? — investigates recovery link by link (setup, connections, WhatsApp, automations and funnel), points to the first problem and tells you how to fix it.
Review my agent’s prompt — looks at the agent’s performance and the reasons conversations were passed to your team, reads the agent’s instructions and suggests concrete edits for each field. You review them and apply the ones you want in the AI Agent menu.
Review my sales pipeline — for CRM-mode stores: where deals pile up or get lost, which open deals are stalled, overdue or worth the most, how the conversations behind them are going, and the three deals or fixes to tackle first.

Assistants that support prompts show these prompts in their prompt or attachment menu. If yours doesn’t, just ask in your own words, as in the examples below.

Plan your next broadcast with AI

Instead of building a broadcast blindly, ask your assistant to analyze your results and recommend the next broadcast. It pulls together what has already worked in your store and suggests:

the audience — which filters to use and how many people that reaches;
the text — what to keep or change in the message, within what Infynia’s template creator supports;
the day and time — when your audience replies and converts the most;
the estimated cost — what the send should cost before you decide.

Then you create and send the broadcast yourself in Infynia, with the Broadcasts wizard. Your AI assistant only recommends: it doesn’t create templates, schedule or send messages.

Examples to copy and paste:

“Analyze my broadcasts from the last 90 days and tell me which was the best, who it was sent to and why you think it worked.”
“Plan my next broadcast: which audience, which text, what day and time, and how much will it cost?”
“How many champion and loyal customers do I have? How much would a broadcast to them cost?”
“Which of the templates I used in broadcasts converts best? Help me write a new version.”
“What’s the best day of the week and time to send to my base?”
“I want to win back people who abandoned their cart and didn’t buy. Which filters can I use and how many people would I reach?”
“Do the weekly review of my store and tell me the 3 most important actions.”
“Why isn’t my store recovering carts?”
“Does my WhatsApp number have any quality or limit problem?”
“Why was my message template rejected and how do I rewrite it?”
“Review my AI agent’s instructions and tell me what to change in each field.”

What you need to know

Read-only — the assistant looks at your data and your store’s configuration, including your AI agents’ configuration and instructions, but never sends messages, changes settings or creates anything in your store.
Owners and Administrators can connect: anyone with either role in the store can make the connection and manage the connected apps. If the person who connected stops being an Owner or Administrator, or leaves the store, their connections stop working.
One store per connection — each connection is tied to the store chosen at authorization time. To look at another store, add another connection and choose the store you want. Connecting the same store again from the same app replaces the previous connection; connections to other stores aren’t affected.
Only what fits your store — the assistant only gets the lookups that make sense for your store’s type and plan. If you ask about something your plan doesn’t include, it tells you which plan unlocks it.
The connection renews itself while you use it. If it sits unused for a long time, or if the assistant says it can’t read something it should (such as your agent’s instructions), connect again.
Customer privacy — the assistant works with aggregate numbers and does not receive your customers’ names, phone numbers or e-mails. The one exception is the optional reading of replies to broadcasts (above), which you turn on per connection and which may show names customers wrote.
You stay in control — you can disconnect any time, on Infynia’s AI Connector page or in your AI assistant’s connector settings.

Manage your connections

On the AI Connector page (in Integrations), the Connected apps section lists the apps that can read the store’s data right now, with who connected them, the connection date, the last renewal, and the Reads replies badge when reading broadcast replies is on.

To cut an app’s access, click Disconnect and confirm. The app loses access to the store’s data immediately; to use it again, just connect again. Use Refresh to reload the list.