Pér
Chatting with Pér¶
Conversation is how you work with Pér. You ask a question in plain language, Pér reads your tenant’s data to answer it, and it shows you what it is doing as it goes. When the answer implies work in Amperity, the conversation is also where that work starts.
A question does not have to be restated as a query, then a ticket, then a round of configuration — it stays one thread, and what comes out of it is a plan you can read. Most of this article is the Understand stage of the customer decision loop; the last part of it is the second way into Approve.
What a conversation is¶
A conversation is a persistent thread of questions and answers against your own customer data.
The chat where you worked out why repeat purchase fell in March is still there in June, with everything it established and everything Pér looked at. You can pick it up rather than start again.
Your chats are yours. Your list shows the conversations you started. A colleague does not see them unless you share one.
A chat is named by your first message until you rename it. Pér does not title your chats for you.
One answer at a time. If you send something while a response is still running, Pér says so rather than starting a second one.
A conversation does not have to happen in Pér. Pér answers in Slack and in Microsoft Teams from the same customer data, in the channel where the work is already being discussed. Those surfaces answer only: they cannot make a change and cannot take an approval, so when something needs changing Pér points back to the Pér web app and the change is made there.
For what Pér reads to answer you, and under whose access, see How Pér uses your data. For the chat surfaces, see Pér in Slack and Pér in Teams.
Asking, and following along¶
Between sending a question and getting an answer, Pér shows its working.
Complex questions may take Pér several steps to answer, so Pér narrates as it goes. You can watch as it narrates, but because the work happens on Amperity’s servers rather than in your browser, it will be there when you come back if you opt not to sit and watch.
How a response behaves:
Pér says what it is about to do before it does it. Each round of work opens with one plain sentence naming the step and why it matters, in business terms rather than tables and columns — for example, “First I’ll pull customers who purchased in the last 12 months.” Alongside it you see what Pér is working on at that moment.
The work continues if you leave. You can move to another page, open a different conversation, or reload the browser; the answer is still produced and saved, and coming back shows you what you missed.
Answers are short on purpose. Pér gives the shortest answer that addresses what you asked, and expects you to ask for more. An answer that runs long opens with a one-line version of the conclusion so you can decide whether to read the rest.
Pér keeps going without asking permission to look things up. It stops to ask only when the choice is genuinely yours: which of several things you meant, an ambiguous target, or a change you did not ask for.
Objects are linked. Where Pér names something in your tenant, it links to it, so you can open it in Amperity.
Note
Pér does not name the tools it calls, and it does not describe how Amperity stores or processes your data — those change, and neither is something you can act on. It reports what it found and what it did. Your own sources, destinations, and bridges are named normally, because they are yours.
Stopping a response¶
You can stop a response that is going the wrong way.
Stopping is recorded on the chat, not on your browser tab. It takes effect even if you started the response somewhere else, or reloaded since.
What Pér had already written is kept, marked as stopped, so the thread still reads in order.
A response that runs very long stops itself and tells you to try again.
Important
Stopping a response does not stop a plan. A plan that is running its own steps is stopped from the plan itself — see Approving or rejecting a change.
Asking Pér to build a plan¶
Ask for something that takes several changes in Amperity, and Pér writes a plan rather than making the changes one at a time.
This is the point where a conversation becomes work, and it is the second way a plan gets written — the other is acting on a recommendation. Both produce the same thing, so what you know about one applies to the other.
A card appears in the conversation with the plan’s title and how many steps there are to review. Open it, read the steps, and approve them there.
Important
Writing a plan changes nothing in Amperity. Nothing in it runs until a person approves it.
For what a plan contains and how its steps are approved, see Plans and How a plan gets written.
Attaching a file¶
You can bring a document into the conversation and have Pér work from it.
Most real questions arrive with something attached — a brief, an export, a spreadsheet someone sent you. Pér reads the text out of it and uses it the way your message asks.
What you can attach:
Plain text files, read directly.
Documents —
.pdf,.docx,.pptx,.xlsxand.xlsm— read on the server and turned into text.Older Word and PowerPoint formats (
.doc,.ppt) cannot be read. Pér says so and tells you which format to save as.
The limits, and what happens at them:
One file at a time. Remove the attached file to attach another.
10 MB for a document, 1 MB for a spreadsheet or a text file, and a ceiling on how much text a file can carry however small the file is.
A file over a limit is refused by name, before you send. Nothing is silently truncated, and the message you typed is not lost.
Important
An attached file stays available to later questions in the same chat — but not forever. In a long enough conversation the oldest attachments are dropped, and Pér says which ones by name so you know it is no longer reading them. Attach the file again, or start a new chat.
Every file you attach is also kept as an artifact, so you can open it again later without digging through the thread.
Note
A file is material for Pér to work from, not a set of orders. If it contains instructions addressed to Pér, they are not carried out. See Files you attach.
Choosing how much thinking Pér does¶
You can trade depth against speed. A one-line lookup and a four-table analysis should not involve the same wait.
There are three levels:
Low effort — quick, lighter responses.
Medium effort — balanced depth and speed. This is the default.
High effort — deeper, more thorough thinking.
Note
Your choice is remembered in the browser you made it in, not on your account. On another computer, Pér starts at the default again.
Starting a skill¶
A skill is a packaged piece of work you can start by name instead of describing from scratch.
The hard part of asking an agent for analysis is knowing what to ask for. Starting a skill skips that: it carries its own method, so you get a considered piece of work rather than whatever a cold question produces.
You can start one from the chat input box, by name.
Pér can start one itself when what you asked for clearly matches a skill, and it says which one it is using — it is never a silent change of mode.
Starting one while an answer is still running puts it in the chat input box ready for your message, rather than interrupting.
Which skills you have depends on your tenant.
Working in a conversation¶
To start a conversation
Type your question and click Ask Pér.
To stop a response
Click Stop generation.
To attach a file
Click + beside the chat input box.
Choose Attach a file and pick the file.
Type your message and send it.
To attach a different file, remove the attached one first.
To change how much thinking Pér does
Click the effort control beside the chat input box.
Choose Low effort, Medium effort or High effort.
To start a skill
Click + beside the chat input box and choose Skills, or type
/in the empty box.Choose the skill you want.
For where the conversation sits in relation to the rest of Pér, see The chat panel.