Skip to main content
The AI assistant is a chat that can look things up in your workspace and act on them. Ask it how last week’s campaign did, who your best customers are, or to draft a campaign for a segment and schedule it. It is not a general chatbot with a memory of your account. It has no stored knowledge of your data. Every fact it gives you comes from a live lookup it runs while answering, in front of you.

​
Where to find it

Sidebar

Senderz AI is the second item in the sidebar.

Top bar

The Marketing Agent button opens a fresh chat from any screen.

Command palette

Search for “Ask the Marketing Agent”.

In context

The Ask AI card on a contact profile and the Ask AI about your analytics bar on the Analytics page open the chat with your question already filled in.
You need an AI provider connected first. See AI provider settings.

​
Who can use it

The chat is available to owners, admins, editors and designers. Viewers cannot open it, because the assistant can create and change workspace content and a viewer is not allowed to do that anywhere else either. A few actions are narrower still and are only offered to owners and admins: re-running domain verification, adding a sender address or sender name, and changing workspace settings. If your role cannot do something, the assistant tells you plainly rather than pretending it worked.

​
What it can read

The assistant can look up almost anything you can see in the interface.
Find a contact by email, phone or name. Read their activity history, lifetime spend and past orders. Count how many contacts match a description. Read your tag list, consent breakdown, RFM tiers and predictive scores. Read your saved segments and how big each one is. Read your contact property catalogue and custom fields.
List recent campaigns, filter them by status, and open one for detail. List automations, open one, and see how many contacts are enrolled. Read templates and preview their content. See what is waiting for approval.
Dashboard analytics, today’s snapshot, deliverability, bounce rate, complaint rate, open rate and click rate, top campaigns by any metric, order counts and recent orders.
Your synced products, product categories, how fresh the catalogue sync is, and the products a specific contact recently viewed or added to cart.
Sending domains and their verification status, the suppression list, whether a specific address is suppressed, your SMS Sender names, your quiet hours, and your recent audit log.
SMS credit balance and recent credit spend, your subscription, connected integrations, brand kit and workspace settings. Coupons, A/B tests and popups.
It cannot read secrets. API keys, signing keys, webhook payloads and other workspaces are outside what any of its lookups can reach.

​
What it can do

The assistant can also write. Every action goes through the same code as the matching button in the interface, so the same validation, permissions and compliance rules apply.
  • Draft a campaign, a segment, an automation or a template
  • Create a coupon, which is also created in your connected store
  • Create a contact, or paste in a small batch of contacts
  • Create a list
  • Set up an A/B test on a campaign
Importing a spreadsheet is not one of these. The assistant can create a small number of contacts you paste into the chat, capped at 25 at a time. A real file upload goes through Contacts → Import. See Importing contacts.

​
What it deliberately cannot do

There is no tool for these, so it is not a matter of the assistant declining. The capability does not exist.
  • It cannot send a campaign to your list. It can only schedule one, and the send pipeline still decides what goes out.
  • It cannot delete a campaign, a template, an automation or a media file.
  • It cannot disconnect an integration, revoke an API key, or change your billing.
  • It cannot change your campaign approval policy, your embedded form origins, or your organisation details.
  • It cannot buy SMS credits or spend anything except through a test send you confirmed.

​
It asks before anything costly or destructive

Some actions change or stop live data, or cost money. For those, the assistant runs the action once with no permission to execute. It gets back a preview, shows you that preview in plain language, and only runs for real after you say yes in the chat. These are the actions that ask first:
Creating a draft does not ask, because a draft sends nothing and is easy to delete. The assistant is still told to state what it is about to create before it creates it, so you should see the name, channel, audience and schedule in the same message.
Scheduling a campaign does not ask for a second confirmation, because it is reversible: the assistant hands you back a one-line way to cancel it, and the send is still subject to every rule below.

​
It cannot get around your rules

The assistant is a different way to press the same buttons. It is not a way around them. See Consent, Quiet hours and Shabbat and SMS.

​
Everything it does is in your audit log

Each write the assistant makes is recorded separately from work done by hand, so you can always tell which was which. Open Settings → Audit Log and look for entries marked as AI-initiated.

​
It treats your data as data

If text stored somewhere in your workspace, such as a contact note or a template body, contains something that reads like an instruction to the model, the assistant is told to ignore it, carry on with what you asked, and tell you it saw an apparent attempt. It does not act on instructions found in your own records.

​
Working in the chat

Past chats are listed in the left panel, grouped into Today, This week and Older. You can rename or delete any of them.Conversations are stored in the browser you are using, scoped to your account and the workspace you are in. They do not follow you to another device, and clearing your browser data removes them. Up to 200 are kept.
The plus button next to the message box attaches a reference to something in your account: an email template, an image from your media library, or an image you upload on the spot. Uploaded images are saved into your media library.An attachment tells the assistant which item you mean so it can look up the real one, rather than inventing a substitute.
After an answer, up to three suggested next questions appear under it, based on what the assistant just looked up. Reading your RFM tiers offers “Who is at risk of churning?” and “Build a segment for this group”. Checking your SMS balance offers “Show me recent SMS spend”. Click one to send it.
  • Copy copies an answer.
  • Regenerate runs your last question again.
  • Edit on one of your own messages rewrites it and re-runs from there. The previous answer is kept as a separate version, and arrows let you move between versions.
  • Stop halts an answer while it is being written.
While it answers, each lookup appears as a small labelled chip: “Reading your segments”, “Counting orders”, “Checking SMS balance”. You can see exactly which data it used to reach an answer.A running token count for the conversation sits in the header, so you can see what a long chat is costing.
The lightbulb button in the chat header opens What I’d do this week: a set of proposed campaigns, automations and segments based on your actual workspace. Optionally give it a focus in the goal field.Accepting one creates an ordinary draft and opens it. Nothing is published and nothing is sent. An accepted automation proposal arrives as a real connected diagram on the canvas, not an outline.

​
Limits worth knowing

  • A message is capped at 4,000 characters.
  • The assistant chains several lookups into one answer, up to 16 steps. A very broad question can hit that ceiling and stop with a partial answer. Ask a narrower question instead.
  • A test send goes to at most 10 recipients.
  • A pasted batch of contacts is capped at 25 at a time.
  • Conversations live in one browser only and are capped at 200.
A weak model gives bad answers. Small or low-cost models often fail to call the lookup tools reliably, and then answer from nothing. If replies look wrong or invented, that is usually the model rather than a bug. The chat marks such a model with a Limited model badge and offers a one-click switch to a stronger one. Pick a capable model in AI provider settings.

​
If it says something is missing

The assistant is instructed never to claim a contact or campaign does not exist unless a lookup actually came back empty in that same answer. If it tells you something was not found, it looked. If you believe the record exists, check the spelling of the email address, or ask it to search by name instead.

Senderz AI overview

Every AI feature in the product and where it lives.

AI provider settings

Keys, models, reasoning, usage and disconnecting.

Campaigns

What the assistant creates when you ask for a campaign.

Automations

What it creates when you ask for an automation.