Assistant

The assistant is a chat panel that knows how NeatPractice works. Ask it a question in plain words, such as "which authorizations expire soon" or "how do I add a payer", and it looks up the answer in your clinic's records or in this help site. It can also open screens when you ask it to and draft changes, such as a new patient or a chart note, that you review before anything is saved.

Opening the assistant#

Click the sparkle button at the top right of any page, or press Ctrl+J (Cmd+J on a Mac). You can also open the command palette and choose Open Assistant. The panel opens beside the page you are on, so you can keep working while you read the answer. Drag its left edge to make it wider or narrower. NeatPractice remembers whether you left it open. On a phone-sized screen the assistant covers the page, so it opens only when you click the sparkle button there; shrinking the window never puts it over a form you are filling in.

The assistant appears only if your clinic's plan includes it, your clinic has it turned on, and your role allows you to use it.

What it can do#

The assistant works with the same access you have. It never sees or changes anything your role does not allow, and if you ask for something outside your access it tells you so.

• Answer questions about your records, such as "what is this patient allergic to", "how many units are left on this authorization", "who is on this patient's care team" or "how many hours did Jordan Example work last week". The answer appears in the chat; see Answers stay in the chat.
• Answer how-to questions from this help site, with a link to the page. It matches your words against page titles, section headings and text, so configuring reminders finds the same pages as reminder settings, and it quotes the sections that name the screen and button. When no page covers the question, it says so rather than guessing a screen.
• Open screens when you ask it to, such as a Setup list or the schedule, and open a patient's chart at any section you can see. When you ask for a filter a screen does not offer, the assistant says so instead of claiming it applied it.
• Find patients by name, chart number, phone or date of birth.
• Check work queues: authorizations that are expiring, used up or still waiting on the payer, and upcoming appointments whose coverage is not verified.
• Check the schedule: appointments for a day or a range, open times at a location, and patients who did not show or cancelled late.
• Check documentation: notes that are unsigned, waiting for a co-sign or sent back, and notes, progress reports, recertifications or re-evaluations that are due.
• Explain audit findings when you can review note audits: list open findings (only the AI guideline ones, or only some severities, if you ask), and explain one: the rule's question and criteria, the verdict, the reasoning, the quoted words, what a failure does and any reviewer rating. It reads only; resolve findings on the Note audits screen.
• Check billing: charges not billed yet, charges on hold and why, and rejected claims waiting to be worked.
• Run reports from the report catalog, including for a provider, location, discipline, payer or patient when the report has that filter, and summarize them. Open shows the report with the same parameters the assistant used.
• Read ABA data when your clinic has the ABA module and you can read treatment plans: a learner's programs and targets with their phase and mastery, a program's graph in numbers (latest values, whether it is improving, worsening or steady toward the goal, phase changes, mastery moves and the graph's event lines, labels and breaks), and recent data sessions with trials, behaviors and notes still unsigned. See graphs and data collection.
• Answer ABA practice questions with the ABA module, each within your permissions: authorization utilization and pace, this month's supervision logs met or at risk, sessions whose note is still unsigned, learners due a reauthorization report, and the counts behind the ABA exceptions widget.
• Answer Setup questions, such as how many payers or providers you have set up.
• Draft changes: a new patient, updated patient details, a contact, a chart note, a coverage, an authorization, a payer, a location, a provider, or booking, moving or cancelling an appointment. Bookings run the same checks as the schedule; if one fails, the assistant tells you why and you can book from the schedule instead.

Answers stay in the chat#

A question gets an answer in the chat, and the page you are on stays as it is. The assistant changes your screen only when you ask it to in so many words, such as "open the payers list", "take me to this patient's coverage" or "show me the schedule". When the answer comes from a screen in NeatPractice, its step in the conversation shows a card naming that screen with an Open button, so you can go there when you want to.

For the questions asked most about a patient, the assistant reads the chart directly:

• Clinical summary: allergies with their reaction and severity, current medications, active alerts and the diagnoses on the patient's episodes.
• Coverage: the patient's insurance and each authorization's status, days left, and visits and units authorized, used, scheduled and remaining.
• Care: episodes of care with their providers and physicians, and the care team.
• Visits: recent visits, upcoming appointments and notes not signed yet.

For anything else it searches the same lists and records you can open in NeatPractice, reads the one that answers the question, and keeps the result short: at most 25 rows at a time, with the total and how to see more. It does not read files, prints or exports, Social Security numbers, keys or other secrets. For totals and counts across many records, such as hours worked by a provider, units used or visits by payer, it runs a report from the report catalog instead of adding up rows.

Analytics questions#

Questions about numbers across many records, such as counts, totals, rates, averages or trends ("what was our show rate last month", "which payer had the most denials this quarter"), are answered only from a report you are allowed to run. The answer names the report and the filters it ran with, such as the dates and how many providers or locations it covered, so you can open the same report and check it. The assistant can also explain a metric, such as how a rate is worked out, from the report's own description and columns.

If none of the reports your role can run answers the question, the assistant says so instead of working the number out another way or estimating it. Ask an administrator for access to the report you need. Reports your role cannot open stay hidden from the assistant too.

Every lookup runs with your access, your location choice and your caseload, and is recorded in the access log like any other time you open a chart. If you do not have access to something, or there is no data, the assistant says so plainly.

Using the screen you are on#

The assistant knows which screen you are on, so you can say "this", "these" or "the selected ones". On a patient's chart, "add a note to this patient" means that patient, and "how is this learner doing on labeling colors" means that learner. On a list, it knows the filters, sort and page you picked, how many rows the list has and which rows you selected, so "how many of these are unbilled" or "open the selected claim" works without spelling the records out. It also knows which tab, record or report is open and the parameters you ran the report with.

With each message the assistant gets only record ids and the list settings, never the names or other details shown in the rows, and never the text you typed in a search box: it only knows that a search is on. When an answer needs what the rows show, such as names, statuses or amounts, the assistant reads the screen again with your access, up to 25 rows of the columns you see, and tells you how many rows there are in all. If the list is longer, it suggests a filter to narrow it. On a screen with no list or report, it uses its other lookups instead.

In a long conversation the assistant keeps the details of its latest lookups and shortens older ones, so it may look something up again rather than rely on an answer from much earlier.

Reviewing changes#

When the assistant wants to change something, it shows a Proposed change card listing every value it will save, in full. Click Apply to save it or Skip to leave things as they are. Before it creates a patient, the assistant checks for an existing chart and tells you if it finds a likely match. You can ask to see a proposed change without applying it; the card is the proposal, and nothing is saved until you click Apply.

While a card waits for you, its step in the conversation reads Awaiting your review. A card stays in place if you close the panel or open your saved conversations and come back, and so does a message you started typing.

Anything that leaves your clinic, such as an eligibility check or a batch of claims sent to the clearinghouse, shows a Confirm before it is sent card. Nothing is sent until you click Confirm. A change you skip reads Skipped and an action you cancel reads Cancelled, never Failed.

Changes made through the assistant are recorded in the change history and the audit log like any other change, marked as made through the assistant.

Saved conversations#

Your conversations are saved so you can come back to them. Click the history button at the top of the panel to see them, open one to continue it, rename it or delete it. The newest show first; click Load more at the end of the list to see older ones. A conversation name can be up to 120 characters. Only you can see your conversations. Your clinic decides how long they are kept; after that they are deleted. Click the new conversation button to start fresh.

Choosing the model tier#

Below the message box, the Model tier picker chooses how capable a model answers you. A higher tier reasons better through hard questions; it does not make answers longer. NeatPractice saves your choice with your account, so it follows you to any device.

The picker shows each tier's cost compared with Small. Start with Small and switch up only when an answer needs more reasoning.

Credits#

Each answer uses a small number of your clinic's assistant credits, more on a higher model tier. If your clinic runs low, the panel shows a notice, and administrators can buy more. See Assistant credits.

Stopping an answer#

Click the stop button while the assistant is answering to stop it. The conversation shows Stopped, and the assistant treats that request as cancelled: your next message gets its own answer, and the stopped request is not finished unless you ask for it again. A change you already applied keeps running when you stop: the message box waits until it finishes, and the conversation shows what it actually did.

When something goes wrong#

If the panel cannot check whether the assistant is available, or cannot load a saved conversation, it says so and shows Try again. A saved conversation never opens empty while its messages are still unavailable.

If an answer fails partway, your message stays in the conversation. Click Try again to send it again without retyping it; anything you have started typing in the message box is kept. When the failed answer had already looked something up or saved a change, ask your question again instead, so nothing is repeated.