®Docs

Assistant

Ask Superpractice about your firm's data, attach records and skills for context, talk instead of typing, and have it take the next step for you.

Open the assistant

There are two ways in:

  • Ask Superpractice, at the bottom right of every page, opens the assistant in a small window over the page you are on. It can see which page you are reading.
  • Assistant, in the sidebar, opens the full page with your conversation history. A grey count beside it shows how many of your chats have an answer you have not opened yet.

In the small window, New chat starts over, Open as full page moves the chat to the Assistant page, and the chat's title opens your history.

Ask a question or give a command in plain words, such as "Who are my hottest leads?" or "What's on my calendar today?". Type @ to mention a contact.

Add records to a message

Click + in the composer to choose what the assistant should look at:

  • Mention a contact
  • Add task context
  • Add contact context
  • Add meeting context
  • Add matter context
  • Skills

Search, or pick from the suggestions, such as today's meetings or tasks due soon. You can add up to 8 records to one message. Each one shows as a chip in your message. Click a chip to open that record.

The assistant reads each record from your firm's data when it answers. Contact notes that are hidden from AI are never included.

Use a skill

A skill is your firm's written way of doing something, such as an intake call or a follow-up after a meeting. Click +, then Skills, and pick one under YOUR SKILLS, or under SUPERPRACTICE SKILLS if your firm has any. You can add up to 3 skills to a message. Each shows as a chip, and the assistant follows it in its answer.

You don't have to attach a skill. When a request matches one of your published skills, the assistant uses it on its own. Either way, the steps it shows include Used skill with the skill's name. To read a skill's text, open it on the Skills page.

A skill can name the other skills it goes with. The assistant gets that list with the skill and reads each related skill when the work reaches it, so it follows "Intake call" and then "Conflict check" in the order the skill says. See Related skills.

Reply to an earlier message

Click Reply under one of the assistant's messages. A Replying to Assistant row, with the message's first line, sits above the box; click × to drop it. Once sent, a short quote sits above your message, and clicking it jumps back to the original.

Get help with a task

On an open task, click Work on task. A new chat starts with the message "Help me with this task", and the task's steps stay pinned at the top of the chat: the task name, how many steps are done, and a progress line. The steps start folded; click the arrow to open them, and tick any step right there. In the popup they sit under the header. The assistant works the steps in order, ticks each one it finishes, and puts what it found or wrote in a note on the task, so everything about the task stays on the task. It never creates other tasks while working one. For steps only you can do, it gives you a checklist or a draft and offers to take the next step. It never marks the task done by itself.

Skills linked in the task's details come along: the assistant follows each one as it reaches that part of the work. See Skills in tasks.

Once you have a chat for the task, the button reads Open Assistant and reopens it. Each person has their own chat for a task. See Tasks.

Attach files

Click Attach files, or drop files onto the composer. The assistant accepts:

  • pictures: JPG, PNG, HEIC, and WebP, up to 4 at a time
  • PDFs
  • Word, Excel, CSV, and text files

Each file can be up to 4 MB, with up to 10 files in a message. The assistant looks at pictures and PDFs. From Word, Excel, CSV, and text files it reads only the text. The original file is kept privately with the chat, so a task you create from the message gets the file itself. Files with macros are refused.

Send more while it is answering

You can keep typing while the assistant works.

  • Enter queues your message. Queued messages wait in a tab above the composer, where you can edit, copy, or delete them. When the answer finishes, everything queued is sent together.
  • Steer, or ⌘Enter, sends a queued message now. It stops the current answer and redirects it.
  • Stop, or Esc, stops the answer. Your queue stays.

While a new message is waiting, nothing that reaches outside the firm is sent. Any card asking for your approval says it is reading your new message first.

Talk instead of typing

Click the voice button, or press ⌥Space, to talk to the assistant. The bar shows whether it is listening, thinking, or speaking, with Mute, Type instead, and End.

  • The conversation keeps going as you move between pages, and between the small window and the full page. Reloading the page or closing the tab ends it.
  • If the connection drops, it reconnects with everything said so far. If it can't, the chat keeps everything that was said, so you can carry on typing.
  • When you ask for a table or a chart, it puts it in the chat and tells you it is there.

Read the answer

  • While it works, the assistant shows a short trail of what it is doing. When it is done, the trail folds to a single line. Click it to see the steps.
  • Charts and tables have an Open larger button. Larger tables can be sorted, and names in them open the contact, matter, task, meeting, or call.
  • Charts that compare two periods name both in the title, such as "Oct 1–9 vs Sep 1–9".
  • Links such as Open in Mission Control open that page in a new tab. Links to a record open the record's window.
  • When it creates or changes a task, a line such as "Task created" with the task's title confirms it was saved. Click the title to open the task. A new task's card lists its steps.
  • A task it creates is attached to the contact and the matter it is about, whatever the task type. If two contacts fit the name, it asks which one you mean instead of guessing.
  • If the answer depends on something only you know, it asks one question with a few options and marks the one it suggests. You can also write your own answer.
  • After it finds options or writes a draft, it offers to take the next step itself, such as booking or sending. It does not act until you say yes.
  • Ask for your booking link ("send Jordan our booking link", "what is our scheduling link") and it reads your meeting types, picks the one whose description fits, and shares that meeting type's booking page in the message or the answer. The same works in the voice assistant and the executive assistant. Internal meeting types are never sent to a contact.

Approvals

Before the assistant does something that can't be taken back, it shows a card with what it is about to do and asks you to Approve or Cancel. Replying that you agree, in your own words, such as "yes", "go ahead", "send it", or "looks good", approves the one card that is waiting, exactly as it shows on screen, including your edits. Your reply is read in the context of the card: a question about it, or a reply that could be read two ways, approves nothing, and the card keeps waiting. Typing "no" or "cancel that" cancels it, and the card says Cancelled. Nothing was sent. If more than one card is waiting, it asks which you mean.

Emails and texts to contacts. In the app, the assistant never sends an email or text to a contact on its own. It shows a draft card: check the words, then click Send email or Send text, or type yes. An email or text set up to go later shows Approve email or Approve text with when it goes. You approve it once, and only those exact words go out at that time. A text still waits for the contact's texting hours, and nothing goes to a contact who opted out.

The card asks which contact you mean first, showing the full number or address, when:

  • the name you said was only a close match, or matches two contacts
  • another contact has a similar name, such as Jon and John Hale
  • the contact was added today
  • the number or address is not the one on file, or not the one your last message to them went to
  • you did not name the contact yourself

After a text is sent, its card says Sent to …, then Delivered to … once the carrier confirms, or Not delivered: with the reason. The assistant tells you about a failed text first in its next answer.

By voice, it puts an email's draft card in the chat and offers to read it to you, then asks for your go-ahead. A bare "okay" or "right" while it is talking only shows you are listening, so it asks again; say yes or tell it to go ahead. Say "no, just send it" to skip the reading. A spoken change makes a new draft. Payments, deletes, sends to many contacts, and commitments made for the firm always ask.

The email draft card

  • To and CC take contacts and teammates, grouped under Contacts and Your team. A full address that isn't a contact shows Not in your contacts with Add as a contact. Send email stays off until To has a contact.
  • Regenerate writes a new draft. Hover over it, or tap it, for Steer: add a note such as "Shorter, and ask about Thursday instead", and click Regenerate. The note shows on the card but is never sent. Steer works on text drafts too.
  • The arrow beside Send email opens Send later: Later today, Tomorrow morning, Monday morning, or Pick date & time, in your own time zone. A scheduled email shows Scheduled with the time, plus Change and Cancel. If it can't go, the card says Not sent: with the reason, and the bell tells you.
  • Discard throws the draft away.

SEO in the assistant

The assistant can check whether keywords are tracked, add them to the Keyword Tracker, read your rankings and AI visibility, and add a competitor, which starts the same sync as the Competitor Intel page.

It always asks before paid work, even when you asked for it. Paid work includes writing, rewriting, or optimizing an article, refreshing a page, keyword and domain research, backlink scans, AI visibility scans, site audits, and adding or analyzing a competitor. The card says what it will do and that it uses paid research or AI. Reading tracked keywords, rankings, and AI visibility that is already measured costs nothing and does not ask.

Web browsing

The assistant can look things up on the web, and work through a website for you, such as finding times for a booking. A live view of the page and screenshots appear in the chat, and only you can open them. For a long task, the chat says it is still browsing, and the result arrives in the same chat. Click Stop to end it.

It can also tell you whether a page has a chat widget, a booking or contact form, a tracking tag, or another script installed, because it reads the page's source as well as its text. Ask with the page's address, for example "Is the chat widget on example.com/contact?"

It never sends a form on a website until you approve it on a Send this form? card.

Turn it off for your firm. Owners and admins can switch off Web browsing and search under Settings > AI Personalization, in Your firm. It is on by default. When off, your assistants never search or visit websites, and nothing is sent to search or browsing services. Company research, SEO data, and syncing your own site are not affected. The assistant's steps then say Web search is off for your firm or Web browsing is off for your firm. See Data retention & AI privacy for what is shared when it browses.

Have an agent call a contact

Ask the assistant to have your firm's voice agent call one of your contacts, for example "call Dana and find a time this week". It shows a card naming the contact, the number and what the call is about, and calls only after you approve. When the call ends, the result comes back to the chat and to your notifications.

Calling hours. A call reaches a contact only inside the hours the law allows calls where they are, on their own clock. Outside them, the assistant tells you before asking for your approval, with the next allowed time in your time zone. It then offers what it can do instead, such as calling at that time or sending a text. See When procedures text and call for the hours in each state.

Call a business for you

Ask the assistant to call a restaurant, shop, or office for you, for example to ask a question or make a reservation. It shows a card naming the business and calls only after you approve. The call is placed by your firm's voice agent that makes outbound calls, to US numbers only. It never uses this to call one of your contacts. The result comes back to the chat.

With web browsing and search off, it can't look up a business's number, but it can still call a number you give it.

If no agent places outbound calls, the assistant says so. An owner can turn on outbound calls for an agent under Agents. See Let the agent place calls.

When you have left the app

If an answer finishes after you have left, the assistant tells you once per chat:

  • a notification in the bell, "Your answer about … is ready"
  • a browser notification, if you turned it on

To turn on browser notifications, open Settings > Notifications, find Browser notifications for assistant answers under Assistant, and click Turn on. It is off by default, is set for each browser separately, and shows only the chat's name. On iPhone, add Superpractice to your Home Screen first. Play sounds in the same section plays a soft sound when the assistant reacts to a message.

Your time zone

"Today", "yesterday", and "this month" in the assistant's answers follow your own time zone. Firm dashboards and reports still count days in the firm's time zone.

Delete a chat

In your conversation history, right-click a chat and choose Delete, then Delete forever. The chat, its messages, and the files you added to it are removed permanently. Tasks it created, messages it sent to contacts, and saved memories stay.

On this page