Provider Manual · Part XV

AI in Hero EMR

Where the AI lives, what it reads, and what you must verify: voice commands, clinical knowledge, product Q&A, note generation, dictation, prescription auto-fill, inbox drafting, and the magic-wand quick actions.

8 sections~17 min read11 screenshots

Part XV

AI in Hero EMR

15.1Voice assistant (F1)

Run hands-free commands with the voice assistant

A round magic-wand button floats at the bottom-right of every Hero EMR screen; hover it and the tooltip reads F1 voice · F3 chat. Press F1 anywhere (or click the wand) and speak. The assistant reads two things — your words and the patient whose chart is open — and turns them into a draft: every command ends in a pre-filled modal or composer for your review, never a direct action.

Start and stop recording. Press F1 or click the wand. The button turns red with a stop square, a waveform animates, and a dark status bar expands to the left; press F1 or click again to stop, or let it auto-stop at 60 seconds. First use asks for browser microphone permission, and if a note dictation is already running you’ll see Stop the active note dictation before starting Command Hero. — the two share one recorder.
Watch the status bar. It shows Transcribing..., then Processing command..., then your transcribed words in quotes with a summary of what the assistant did underneath. Results clear themselves after about five seconds of inactivity.
Confirm the patient. With a patient chart or patient tabs open, the command applies to that patient. Otherwise the assistant opens the patient search modal and shows Please select a patient to continue... before doing anything.

You say	What opens	What’s drafted from your words
Prescribe a medication	`Generating prescription details...`, then an order modal titled `Prescription from Command Hero`	A fully drafted sig, built by the same engine as Prescription Auto-fill.
Order a lab	`Lab Order from Command Hero`	The test, matched against the lab compendiums your practice has enabled (see Orders); if nothing matches you get a `No enabled lab compendium match found` message naming the test.
Order imaging	`Radiology Order from Command Hero`	Modality, body part, and laterality — “right knee” sets laterality R.
Send a portal message	The `New Message` composer (see Composing messages)	Recipient, subject, and body, pre-filled for your review.

A patient chart with the floating magic-wand button in its active listening state at the bottom-right — a red recording pill with a live audio-level indicator — Press `F1` and the wand flips into a red recording pill with a live audio-level indicator.

Clinical questions work here too: ask one aloud and instead of an order you get a guideline-backed answer panel — see Knowledge Hero.

The assistant drafts; you approve. Review every pre-filled field of a voice-drafted prescription, lab, imaging order, or message before you save or send it — nothing is transmitted until you do.

15.2Knowledge Hero

Ask clinical questions with Knowledge Hero

Speak a clinical question to the F1 voice assistant — first-line treatment, contraindications, diagnostic criteria — and Knowledge Hero answers it from your practice’s guideline library. The status bar turns emerald with Searching clinical guidelines..., then a panel slides in from the right titled Knowledge Hero, subtitled Clinical Decision Support. There is no typed entry point — Knowledge Hero is voice-only (the F3 chat answers questions about the software, not clinical ones).

Read the answer and its sources. The panel shows Your Question, the answer, and a numbered References list; header badges count the sources used (“N guidelines referenced”). Afterwards the status bar reports Found answer using N guideline(s) or Answer generated (no specific guidelines matched) — weigh an unmatched answer accordingly.
Copy or close. Copy Answer (it flips to Copied!) puts the answer on your clipboard; Done or Esc closes the panel.
Bring in patient context deliberately. Chart data is included only when your question references the patient (“this patient”, “their history”) and a chart is open — the panel then carries a Patient context included badge. General questions are answered without chart access, so phrasing matters when you want a chart-aware answer.
Curate the guideline library. Open the Settings gear and choose Knowledge Hero Guidelines. On the Guidelines list you can enable or disable individual guidelines, add and edit custom ones, Revert to Original, or Delete Custom Guideline. The Settings tab holds one toggle — Enable previsit workup generation based on previsit interview? — which makes completed previsit interviews generate Knowledge Hero guidance automatically (see Questionnaires & paperwork).

Decision support, not a decision. Answers are drawn only from the guidelines your practice has enabled and are no substitute for clinical judgment — check the References list before acting on one.

15.3Product Q&A (F3)

Ask how Hero EMR works (F3 chat)

Press F3 anywhere to toggle a compact chat panel above the wand button, titled Codebase Q&A with the subtitle Ask how the app works · F3. Use it for how-do-I and where-is questions about Hero EMR itself — workflow setup, feature locations, configuration. It is not a clinical reference, and it does not read patient data.

Open it. Press F3 (you must be signed in). Dismiss it with Esc or the Close (Esc) button.
Pick an example or ask your own. The empty state offers three clickable starters — How do I set up a cash-pay only visit type?, Where is insurance eligibility checked?, How does proxy (minor) portal messaging work? — plus the hint Ask about how the Hero EMR codebases work or where a feature lives.
Send and read. Type into Ask a question… and press Enter (Shift+Enter for a new line). The panel shows Searching the codebase… while it works, then answers with headings, bullets, and numbered steps; any errors render inline in the thread.

The Codebase Q&A panel docked at the bottom-right with the subtitle Ask how the app works · F3, three sample question buttons, and an Ask a question input — The `F3` panel’s empty state — three example questions and the `Ask a question…` input.

Three F-keys, three jobs: F1 is the voice assistant, F3 is this product chat, and inside the Note Writer F2 jumps to the next template field (Shift+F2 goes back). They are unrelated.

Answers are generated. They describe current product behavior in plain language, but if an answer ever conflicts with this manual, trust the manual — and let your Hero EMR contact know about the discrepancy.

15.4Note Writer AI

Generate note content with AI in the Note Writer

Two AI features live inside the note editor. AI Note Generation answers questions about the open patient while you write, via the .hero dot-phrase; Smart Encounter Completion extracts structured data from your note when you save. Both read the open patient’s chart and your note text — and both end in content you review before it counts.

Generate with .hero. Type .hero in the editor followed by a clinical prompt. The dropdown reads Type or select a question about this patient... and offers quick examples such as Summarize patient’s cardiovascular history, List all previous surgeries, and Summarize recent lab trends; your favorites and recent prompts are remembered. .hero is patient-bound — it reads the open chart, unlike Knowledge Hero, which answers general clinical questions.
Know which dot-phrases use AI. Among the data dot-phrases, only .imagesum is AI-powered — it inserts an AI imaging summary (search prompt Search imaging summaries). The others (.lab, .imaging, .questionnaire, .media, .derm) insert chart data directly, with no model in between.
Keep the built-in guide handy. The toolbar Info button (Note Writer info) opens the Note Writer Guide — tabs Shortcuts, AI & Data, Dictation, Collab, and Workflow — the in-app reference for everything in this section. (The Shortcuts tab also covers F2/Shift+F2 template-field navigation, which has nothing to do with the F1/F3 assistants.)
Save, then review the extraction. Saving the note triggers Smart Encounter Completion — as the guide puts it, “When you save, AI automatically extracts structured data from your note”: Diagnoses & ICD-10, Medications, Billing codes, Follow-up plan, and After-visit summary. A processing modal shows Processing..., then AI extraction complete and Encounter completion data has been refreshed. Everything lands in the Encounter Completion checklist for review — see Auto-completion.

Verify before you sign. Extracted diagnoses, billing codes, medications, and follow-up are proposals — the Encounter Completion review is where you confirm or correct each one before Sign & Close Visit.

Not everything that suggests is AI: the E/M Code Suggestions panel in charge capture (with its Recommended badge) is rule-based by visit mode. The AI contribution to billing is the save-time extraction above — see Charges & E/M.

15.5Dictation & transcripts

Dictate notes and capture telehealth audio

The round microphone button in the note header and toolbar records the visit. When you stop, the AI reads the transcript and your active SOAP template and generates a full note that streams into the editor — structured, templated prose, not your words verbatim. For telehealth, an audio-settings panel lets the same recording capture the Zoom side of the conversation.

Record, stop, and watch it generate. Click the mic (Start Recording / Stop Recording). The status runs Live → Transcribing → Generating note… → Done as the note streams in. The mic and the F1 assistant share one recorder, so they can’t run at once; if nothing is heard you’ll see a No speech detected countdown before recording stops on its own.
Capture Zoom audio for telehealth. The chevron next to the mic (Show audio settings) opens the Audio Source panel: a Capture Zoom audio (virtual device) checkbox, a Microphone picker, and a Zoom / System Audio Input picker (Select a device (recommended: ZoomAudioDevice (Virtual))). The help text covers a BlackHole + Multi-Output fallback and warns: “Select a virtual audio input above to include Zoom audio in the recording.” Every computer’s audio setup is different — test yours before a live visit. Connecting and running Zoom visits is covered in Run telehealth visits.
Turn on speaker labels if you need them. Settings gear → SOAP Notes → Settings → Note Writer Preferences → Speaker Diarization. As the setting itself says: “Leave this off unless you specifically want speaker-labeled transcription.”
Navigate away safely. Leaving the note mid-dictation shows a compact dock at the bottom-right with the patient’s name, a Live/Processing/Done badge, and Complete record / Return to note actions. The browser also warns before you close the tab while a dictation is in progress.

Connection hiccup? Recordings are cached locally. A Retranscribe from cached recording menu appears next to the mic when cached sessions exist, and after a connection problem an alert offers Retranscribe now / Later — you don’t lose the audio.

Proofread before you save. Stopping the recording generates a templated note from the transcript — expect generation, not a verbatim transcript, and read every section before saving and signing.

15.6Prescription Auto-fill

Auto-fill prescriptions from your note (Medication Hero)

Wherever you build a prescription — encounter completion orders (Prescribing) or the standalone order flow (Standalone orders) — the Edit Medication dialog has an orange Auto-fill button (wand icon, tooltip Autofill using clinical context from your SOAP note). It sends the medication, your SOAP note, and patient context to Medication Hero, shows Loading..., and drafts the whole regimen. The same engine powers the F1 prescribe command (Voice assistant) and order extraction from faxes and inbox messages (Faxed orders, Inbox AI).

Document the plan first, then click Auto-fill. Auto-fill quality depends on what’s in your note. Run it after the plan is written and the fields are derived from your documentation; run it before documenting and you get defaults, not chart-derived values.
Review every filled field. It fills action, dose and dose units, route, frequency, duration, dispense quantity and units, and a recommended NDC, and it can set the PRN state — ticking As needed (PRN) and filling the reason field (As needed for... (e.g. pain, nausea, anxiety)). The regenerated sig preview mirrors exactly what the pharmacy will receive — read it the way the pharmacist will.
Read the warning banners. A Medication Hero warning amber banner lists any safety flags from the analysis; if key values couldn’t be determined you’ll see Some key information could not be determined, please check medication dosage and duration carefully. Controlled substances also show a PDMP status box — see Safety checks.
Controlled drugs stay structured. For Schedule II–V medications the Use Free Text Prescription toggle is disabled (“Free-text prescriptions are disabled for Schedule II-IV medications. Use the structured prescription fields instead.”), so AI suggestions land in discrete, reviewable fields — nothing hides in free text. Signing them uses EPCS.

The Edit Medication dialog after Auto-fill with structured dose, dispense, frequency, and duration filled in and an amber Medication Hero warning giving indication-specific dosing guidance — `Auto-fill` fills the structured sig and surfaces a Medication Hero warning with indication-specific dosing to check it against.

Verify dose and duration yourself. If the analysis finds nothing usable, Auto-fill falls back to generic defaults (take 1 tablet by mouth daily for 30 days) — never assume the values came from your note. The warning banners are advisory and do not block Save Changes.

Order entry only: Auto-fill exists in the order-entry dialog (the one that ends in Save Changes). The renewal version opened from the Rx inbox via Approve with Changes has no Auto-fill button — see Refill requests.

15.7Inbox AI

Use AI in the Inbox: drafted replies & order extraction

Four Inbox surfaces carry an AI assist, and all of them follow one pattern: the model drafts, the draft lands in a composer or panel, and you edit and send. The click-by-click reply workflow lives in AI-assisted responses; this section maps where each control is, what each model reads, and what to check before sending.

Surface	Where the control is	What the model reads
Patient messages	The purple wand (`Generate AI responses`) under `Your Response to Patient`; disabled with `This message is not linked to a conversation thread` on unthreaded items	The conversation thread and chart context, steered by the optional instruction box (`Optional: Tell AI how to respond (e.g. 'reassure, schedule follow-up in 3 months')`).
Order extraction	A small wand (`Extract orders`) on each of your sent messages in a thread	The text of your sent message — the instructions you wrote to the patient, not the patient’s message. Phrase instructions completely if you plan to extract from them.
Result replies	The `AI` wand button in the results reply composer	The result and chart context, with the same optional instruction input; the chosen draft is marked `Using AI Response #N`.
Website inquiries	`Generate` under `AI Suggestions` on an inquiry	The contact-form inquiry, auto-analyzed on arrival into an `AI Analysis` card with Category and Summary.

Generate and pick a draft. On messages and results, an instruction gets you one tailored response and no instruction gets two alternatives, opening in the AI Suggested Responses modal as Response 1/Response 2 cards with a one-line rationale. On inquiries, the draft appears under AI Suggestions on the inquiry itself and a toast confirms Inquiry Hero drafted a response.
Edit, then send it yourself. Clicking a card drops it into the composer. Nothing is ever sent automatically — you press Send to Patient (messages and results) or Send Response (inquiries) after editing. A closed inquiry shows This inquiry is closed. Reopen it from the status dropdown to send a response.
Treat extracted orders as drafts. Extract orders parses your message into draft diagnoses, medications, labs, imaging, referrals, questionnaire orders, and a follow-up timeframe in collapsible panels with counts (for example 3 orders). It needs the thread linked to an encounter (This message is not linked to an encounter) and the message fully sent (Wait for the message to send before extracting orders); the drafts then go through the normal order review and safety checks — see Safety checks.

The Inbox Patient view with a thread open: Message History with a clinician reply bubble, and the Your Response to Patient composer carrying a Response 1 AI chip above the Send to Patient button — An AI draft loaded into `Your Response to Patient` — the `Response 1` chip marks its origin; you still edit and send.

A CT Pulmonary Angiography result with a CRITICAL badge, the report text, and the Your Response to Patient composer with its AI button and Send to Patient — The results composer has the same pattern — the `AI` button drafts a patient-friendly explanation of the result.

Every draft is a suggestion. Read it as the patient will — verify the clinical content, the tone, and any commitments it makes (follow-ups, refills, reassurance) before you press send. The drafts read the chart, but you are accountable for the reply.

Your edits teach the model: when you modify an AI draft before sending, the change is recorded to improve future suggestions — worth knowing if your practice documents its data-handling practices.

15.8Magic wand quick actions

Reach Command Hero from the magic wand — hover, type, or click

The round magic-wand button at the bottom-right of every screen is more than the F1 voice trigger. Hover it and a vertical flyout of quick-action buttons rises above it, each showing its hotkey in a tooltip. Clicking the wand itself still starts and stops the voice assistant, and F3 still opens the product Q&A chat — the flyout simply adds typed and click entry points to the same command engine, so you never have to speak to use it.

The magic-wand button hovered on a patient chart, showing a vertical flyout of three round quick-action buttons stacked above it and a dark tooltip reading Type a command · F2 — Hovering the wand reveals the quick-action flyout; each button names its hotkey on hover — here `Type a command · F2`.

Type a command — F2. Press F2 (or click the keyboard button) and a small Type a command for Hero box opens just above the wand. Whatever you type runs through the same pipeline as the voice assistant — “order a CBC”, “prescribe amoxicillin 500 mg”, “message John Smith that his labs look fine” — and ends in the matching drafted order, prescription, or message for your review. Enter sends, Esc closes, and the dark status bar to the left of the wand shows Processing… and the result exactly as it does for voice. The footnote says it plainly: Same commands as the voice assistant · Enter to send · Esc to close.

The F2 box takes the same phrasing you would speak — type it, press Enter, and review the draft it opens.
Create orders — Ctrl+O. The clipboard button (or Ctrl+O) opens the standard order-entry dialog, Create Orders, for medications, labs, imaging, screeners, and forms. With a patient chart open, that patient is already selected — the dialog carries their name in its title; with no patient in context it opens the patient search first, exactly like a voice-driven order. This action is hidden for staff roles.

Opened from a chart, Create Orders arrives with the patient preselected — the same dialog the voice assistant and encounter Orders use.
Quick note — Ctrl+N (physicians only). The sticky-note button (or Ctrl+N) opens the Quick note dialog: a rich-text editor with a Bold, Italic, Underline, and bulleted/numbered-list toolbar. Its subtitle spells out where the note goes — … files to the chart under a new notes-only encounter. As with orders, no patient in context opens the patient search first. Quick note is available to physicians only.

The Quick note editor with formatting applied — a bold heading and a bulleted plan.

Dot phrases work here just like the Note Writer — type a period and a few letters to open the phrase menu.
File it, and it is on the chart. Sign & file to chart creates the notes-only encounter, signs the note, and confirms with a green Quick note signed and filed to the chart. banner — filing is signing, with no separate step.

Filing signs the note and drops it onto the chart under a new notes-only encounter.

Windows and Ctrl+N: Chrome reserves Ctrl+N for a new browser window on Windows, so that shortcut may not reach the app there — hover the wand and click the Quick note button instead. The flyout button always works, on every platform.

Quick actions wait their turn. Every wand action stays inert while another dialog is open, so a stray F2, Ctrl+O, or Ctrl+N can never stack a second window over work in progress. And as everywhere in Command Hero, typed commands draft orders, prescriptions, and messages for you to review — nothing is transmitted until you confirm.

Need help? Email support@heroemr.com.