TPF Resources · AI documentation · Free · Open source · CC BY 4.0

Paramedic-Narrative: free AI help for clearer patient care report narratives

Paramedic-Narrative is a free tool from The Paramedic Foundation (TPF) for paramedics and emergency medical technicians (EMTs). It turns the call details you give it into a draft patient care report (PCR) narrative in your agency’s format. It asks for what is missing. Anything still unresolved stays marked for you to check, instead of being guessed at.

14
Narrative formats
SOAP · CHART · DCHART-E · 11 more
4
Clinical role contexts
Emergency · Rescue · Community · Hospital
17
Knowledge files
Named for the situation they apply to
0
Clinical decisions made
Editorial tool only · Free · CC BY 4.0

Editorial tool only

Paramedic-Narrative helps you write the narrative. It does not assess patients, interpret findings or recommend treatment. It must not be used to inform a clinical decision of any kind. Every draft requires your review, and you keep full professional and legal responsibility for what you submit. The Paramedic Foundation makes no warranty as to the accuracy of any output. Read ETHICS.md before use →

Download for Claude → Choose another platform

Free, no registration, version 3.0.0, Creative Commons Attribution 4.0 (CC BY 4.0) license. Works on desktop and phone. Source and releases on GitHub. You are responsible for complying with your agency’s policy on AI tool use.

On this page

What it does for you

What Paramedic-Narrative does

It spends less of your time building the narrative and leaves more for checking it. It focuses on the parts of charting that take longest and carry the most risk: structure, recall and spotting what is still unconfirmed.

01
A draft from the details you give it
Describe the call by voice, by typing, or by photographing clinical data. The draft is assembled from what you supplied, and it does not invent, assume, or infer a vital, dose, finding, or time.
02
Your agency’s own format
Fourteen formats, SOAP with a Clinical Summary by default. Your agency declares the format, and you can switch it for a single call.
03
Only the questions that matter
Prompts ask only for what the narrative needs and you have not already said. Anything that belongs in a structured field of the electronic PCR (ePCR) is referred to, so you do not type it twice.
04
Unresolved information stays visible
Anything uncertain, computed, or unconfirmed is marked [VERIFY] and listed before drafting, so gaps reach you as a checklist rather than as plausible-looking text.

How it works

Three steps, and you stay in control of all three

Step 1
Describe or dictate the call
Talk it through in any order. Fragments are fine, and you can add to a call across a shift. Photograph monitor screens, vials or paperwork once identifiers are removed. The tool copies values as shown and confirms them with you. If you chart hours later, you get pointed questions to help you recall, instead of a fresh interview.
Step 2
Review the questions, then the draft
Before drafting, you get one grouped list of what is still open: missing items, conflicts, unclear attribution, and any value that was calculated rather than supplied. Answer what you can, then the draft is written in your format.
Step 3
Verify, approve, and paste into your ePCR
Check every [VERIFY] item, confirm the clinical characterizations are yours, then copy the text into your ePCR narrative field. The tool has no connection to any ePCR platform: it cannot read your chart or submit anything for you.

The 12-point dictation skeleton comes as a pocket card, a wallet card, and a phone image to keep open while you dictate.

What it covers

The documentation standard is inside the tool

The tool asks for what a narrative must carry and structured fields cannot hold. Those items fall into four areas.

01
Narrative structure and completeness
Fourteen formats, including SOAP, CHART, and DCHART-E. History source and reliability, scene observation, pain assessment, protocol and clinical practice guideline (CPG) naming, age-stratified vital sign ranges, and validated scoring instruments.
02
Medication, controlled substance, and forensic detail
Indication, dose rationale, weight-based calculation, response, and anticipated effects kept distinct from adverse events. Complete controlled substance audit trail elements. On forensic calls, source attribution for every claim, verbatim quotes, and observation separated from inference.
03
Refusals, handoff, transfer, and late charting
Consent, capacity, and medical necessity on refusals, cancellations, and low-acuity calls. IMIST-AMBO handoff prep and prearrival notes on request. Delayed and fragmented charting, with recall gaps left marked rather than filled.
04
Attribution and agency requirements
A structured ePCR entry asserts that your crew performed that act, and feeds external reporting as such. Care by another agency, care given before you arrived, and anything prepared but not performed stay out of those fields and go into the narrative, attributed to whoever acted.
Full reference lists: formats, scoring instruments, and specialized content

Narrative formats (14)

SOAP with Clinical Summary (default), SOAPE, SOAPIER, CHART, CHARTE, DCHART-E, chronological, head-to-toe, DRAATT, AT CHART, FACT, refusal, interfacility transport, and agency-defined formats.

Scoring tools and decision instruments

SALT triage, ACS Field Triage, HEART score, Cincinnati Stroke, NIHSS, CIWA-Ar, COWS, C-SSRS, RASS, GCS components, Apgar, Ottawa Rules, AUDIT-C, CAGE, PHQ-2, Clinical Frailty Scale, and others. The tool documents an instrument you applied or considered. It does not score a patient.

Specialized content

  • Postpartum and obstetric risk: a 12-month postpartum prompt for women of childbearing age regardless of chief complaint, with GPAL history and dating.
  • Substance use and withdrawal: last drink, prior withdrawal seizures, prior delirium tremens, medication-assisted treatment.
  • Behavioral health: de-escalation, co-response role attribution, RASS trajectory, alternative disposition.
  • Care pathway and barriers: presenting need against dispatch complaint, access and system barriers.
  • Multi-patient incidents: one workspace per patient, with nothing carried between patients or incidents.
  • Trauma handoff to the Advanced Trauma Life Support (ATLS) 11th edition standard.

What it does not do

  • Make or support a clinical decision of any kind
  • Invent, assume, or infer clinical detail
  • Access an ePCR platform or submit a document
  • Fabricate any part of a controlled substance audit trail
  • Cite a structured entry that does not exist
  • Assume your crew performed an act whose performer is unstated
  • Suggest assessment, treatment, or destination during a call
  • Accept photographs containing identifiable information

Four role contexts

State your role at the start of a session, and the prompts, framework and disposition options adjust to it. If you work for more than one organization, say “switch to [agency name]” partway through. Your provider profile stays active.

Emergency
Emergency Paramedic
911 response. Full standard: scoring tools, forensic documentation, IMIST-AMBO handoff, and care pathway documentation for refusals.
Community
Community Paramedic
Community paramedic (CP) and mobile integrated healthcare (MIH) visits. Visit reason, functional status, medication adherence, resource connections, and care plan, with alternative disposition the usual outcome.
Rescue
Rescue Paramedic
Technical and special operations. Adds scene safety and hazard documentation, rescue mechanism, extrication detail, multi-agency role attribution.
Hospital
Hospital Paramedic
Critical care transport, interfacility, in-hospital response. Transport indication and medical necessity, pre-transport stability, en route interventions, structured handoff.

Release

What changed in version 3.0.0

Version 3.0.0 changes how the standard is stored. What it says is the same: no threshold, criterion or rule changed.

Each platform used to carry its own copy of the standard, cut down by hand from one source. Those copies did not stay in line. An audit before release found twenty problems, and fourteen of them came from that drift. On ChatGPT, the forensic section listed five of its eleven triggers, and the standing disclaimer had been cut short.

There is now a single source. The rules that apply on every call sit in one block, copied exactly to every platform. The rest is split into 17 files, each named for the situation it covers, which helps the assistant find the right part. Each platform’s files are built from that source instead of written separately. A check at build time confirms that the always-loaded rules match and that every file can be reached.

The assistant is never certain to find the right file. Naming files by situation improves the odds, and the always-loaded rules do not depend on it. Technical detail → · Full release notes →

Upgrading from 2.x? The file layout changed. Remove the old SKILL.md, documentation-standards-primer.md, and narrative-formats.md first, so the assistant cannot consult a retired copy. See Updating an existing installation. Provider profiles and agency configurations are unaffected.

For agencies and Chief Paramedics

One configuration file, one documentation standard across your system

An agency configuration file tells the tool how your system writes its charts. It sets the narrative format, protocol titles, ePCR habits, controlled substance rules, handoff expectations and local policy. Everyone who loads it works from the same instructions, instead of their own reading of them.

Configuration supports consistency. On its own, it does not make you compliant with your policies, state requirements or any regulation. It also does not reduce the duty of the person charting to review every chart.

The builder is part of the tool. An administrator says “I want to set up agency configuration” and is guided through each section. The administrator can upload protocol PDFs, controlled substance standard operating procedures (SOPs) and documentation standards, so details are drawn from them. The Chief Paramedic reviews the result and affirms an endorsement statement before it goes out.

Section 4A: structured-field scope and attribution

Section 4A records which structured entries feed external reporting. It records whether your agency holds waivers or variances for particular medications or procedures. It also records how care by partner agencies and care before your arrival are handled. This matters on shared calls. Say two agencies each hold an authorization. If you enter the other agency’s medication administration in your fields, one act is reported twice and credited to an agency that did not perform it. Completing 4A lets the tool apply your rule instead of a general one.

What the configuration covers

  • ePCR platform and narrative field specification
  • Required narrative format, including custom formats
  • Structured-field scope and attribution boundary (Section 4A)
  • Protocol titles by call type and adopted CPG sources
  • Refusal and non-transport protocol reference
  • Controlled substance policy: witness, waste, reconciliation
  • Prompt settings per call type (on, off, or required)
  • Transfer of care standards and receiving facilities
  • Alternate destination authorization
  • Chief Paramedic endorsement and scope
  • Privacy and AI use policy reference
  • Configuration version history

Quick start for administrators

Step 1
Start the template
Download agency-config-template.md, or build it with the tool.
Step 2
Chief Paramedic reviews and endorses
Section 2 carries the endorsement statement, creating a record of Chief Paramedic oversight.
Step 3
Host and distribute
Post the completed agency-config.md on your intranet, shared drive, or ePCR document library for staff to load.
Updates
Keep it versioned
When protocols change, update the file, increment the version, and tell staff. The builder warns before overwriting.

Questions about configuration or AI governance for your system? Contact TPF directly →

Installation

Set up once, then use it on every shift

Every platform gets the same always-loaded rules and the same 17 knowledge files. Only the method differs. The iOS and Android tab covers what cannot be done in an app. Why 17 files →

Claude — skill package, or Files and Instructions

Option A: the skill package. One upload, where your account has the Skills feature.

  1. Download paramedic-narrative.skill from the Releases page.
  2. In your Project, open Project settings → Skills and upload it. The instructions and all 17 knowledge files come with it.

Option B: Files and Instructions. Use this if there is no Skills option, or the upload appears as garbled text.

  1. Create a Project and name it something you will recognize, such as PCR Narratives.
  2. Copy everything below the horizontal rule in claude-project-instructions.md into the Project’s Instructions field.
  3. Upload the 17 WHEN-*.md files from paramedic-narrative/references/ to Files, choosing UTF-8 if asked.
  4. Optional: add your agency-config.md to the same Files section.

Do not paste system-prompt.md into a Claude Project. It is for platforms without file retrieval. Upgrading from 2.x: remove the three retired files from Files first. Provider profiles, agency configurations, and past conversations are unaffected.

ChatGPT — Custom GPT

  1. Copy everything below the horizontal rule in chatgpt-instructions.md. The header lines are not part of the paste.
  2. In ChatGPT, go to Explore GPTs → Create → Configure and paste into the Instructions field.
  3. Download the 17 WHEN-*.md files from paramedic-narrative/references/. Under Knowledge, choose Upload files and add all 17, keeping the filenames as they are.
  4. Name the GPT and save. Your agency configuration can go in as an eighteenth Knowledge File.

Upgrading from 2.x: delete the old SKILL.md from Knowledge before uploading the new files.

Google Gemini — Gem, single paste

  1. Copy the full contents of system-prompt.md.
  2. In Gemini, go to Gems → New Gem and paste into the instructions field.
  3. Name it, paste your agency configuration below if you have one, and save.

Gemini has no file retrieval, so system-prompt.md carries the always-loaded rules and all 17 knowledge sections inline. Nothing else to upload. To update, paste the new version over the old.

Microsoft Copilot — one-time setup plus a per-session paste

  1. Once: copy everything below the horizontal rule in chatgpt-instructions.md. In Copilot, open the three-dot menu → Settings → Personalization → Edit instructions, paste, and save.
  2. Each session: open system-prompt.md, click Raw, select all, and copy. Paste it as your first message with the note: “This is your clinical reference standard for this session.” Then describe your call.

Copilot keeps Custom Instructions between sessions but has no persistent knowledge file feature, which is why the reference content is pasted each session. If your organization restricts Custom Instructions, paste both files each time. Copilot Studio or Agent Builder can hold the reference files in a declarative agent instead, removing the per-session paste.

iOS and Android

All four platforms have free mobile apps and work in a phone browser. Setup follows the desktop steps, with three differences worth knowing.

Getting files
The GitHub app cannot download raw files. Use Safari or Chrome: open the file, tap Raw, then tap and hold to select all and copy. On iOS, the share icon saves it to the Files app.
Web only
Creating or editing a Custom GPT or a Gem is a web feature. Do that in a phone browser, then use the apps to run it.
Shortest setup
Gemini needs one paste and no uploads. On Claude, the skill package is a single upload where Skills is available.

On shift, dictate with the microphone and use the camera for printouts, 12-lead strips, and medication lists once identifiers are covered. Then approve the draft, copy it and paste it into your ePCR app.

Policy notice

The Paramedic Foundation does not condone violation of any agency policy, jurisdictional regulation, employer requirement, or regulatory prohibition on AI tool use. Each user is solely responsible for knowing and complying with all applicable policies before use.

Any large language model (LLM) platform with custom instructions

  1. Open system-prompt.md and copy the full contents.
  2. Paste it into whatever the platform calls its system prompt, custom instructions, or persona field.

That file carries the always-loaded rules and all 17 knowledge sections inline, so no retrieval is needed. It also works as the system message for an application programming interface (API) integration. Platforms with retrieval can instead load the always-loaded block plus the 17 WHEN-*.md files separately, as Claude and ChatGPT do.

Governance, ethics, and privacy

Accountability stays with the paramedic or EMT

The duties below come from the ethical framework of autonomous paramedic practice. Software does not enforce them. Users accept them by using the tool.

“Artificial intelligence may augment paramedic cognition, pattern recognition, documentation, logistics, and clinical decision support. It shall not replace identifiable professional accountability. Every patient care pathway influenced by AI must retain clear and traceable clinical responsibility.”

— Nudell, N. G. (2026). Clinical governance in the age of artificial intelligence: A profession-wide imperative for paramedicine. Governing Care. The Paramedic Foundation.

Responsibility
The chart is your professional document
AI assistance does not transfer or dilute clinical and legal responsibility. If the tool makes an error and you submit it uncorrected, the error is yours.
Review
Review every draft before submission
Verify every [VERIFY] item and confirm that each clinical characterization reflects your own assessment. A submitted narrative is a permanent legal record.
Controlled substances
Audit trail integrity
No element of a controlled substance audit trail, including witness identity, amount drawn, amount wasted, and chain of custody, may be invented, estimated, or reconstructed without direct knowledge. Falsification is a criminal matter, not a charting error.
Hard limit
Never for clinical decisions
This tool documents decisions paramedics have already made. Using it to inform, guide, support, or rationalize a clinical decision is unethical, unsupported by evidence, and may breach applicable licensing and scope-of-practice law.
Privacy
Patient privacy is your responsibility
Do not put patient-identifying information into an AI session. Never photograph a face, a license plate, or a document showing a name or date of birth. Crop or cover identifiers first, and dictate values when an image cannot be fully redacted. Camera metadata can carry location data that is itself identifying. Your agency’s privacy obligations and applicable law apply to anything you enter. The Paramedic Foundation is not responsible for breaches resulting from user conduct.

The full ethical framework is in ETHICS.md.

Technical appendix

How the standard is stored

For readers evaluating or contributing to the project. Nothing here changes how the tool is used.

The assistant does not read uploaded reference files end to end on every message. It breaks them into pieces, searches them and looks only at the pieces that seem relevant. So a rule can sit in a file and still be missed where it applies.

Version 3.0.0 is built around that. Rules that must hold on every call do not depend on that search. They sit in the always-loaded block, identical on Claude, ChatGPT, Gemini and the direct API. The rest is split into 17 files, each named for the situation it covers, such as WHEN-FORENSIC.md and WHEN-NO-TRANSPORT.md. Each file states its trigger on the first line. A router matches situations to file names, and the assistant names the file it consults.

A build script generates the four platform files, instead of each being kept up by hand. The build fails in five cases. The always-loaded text differs by even one byte between platforms. The router cannot reach a knowledge file, or it points at a file that does not exist. The ChatGPT instructions run over 6,500 characters, or the text breaks the house style rules. That check was tested against the defects found in the previous release, which are listed in the release notes.

The cost is a longer setup: 17 files to upload once, instead of two. If the assistant never names a file across several calls, something is wrong, and TPF would like to know. Open an issue.

International adaptation

Built to be adapted to each system

Regulatory frameworks, scope of practice, protocol structures, ePCR platforms and handoff standards differ from one jurisdiction to the next. Support for many formats covers part of that. Systems that chart in CHART, chronological or locally defined structures can use the tool as it ships.

The CC BY 4.0 license lets any paramedicine organization copy (fork) the repository. It can adapt the clinical standards and protocol references to its own system and publish a version for its jurisdiction. The clinical reasoning framework and governance standards carry over. Protocol names and regulatory references can change. Splitting the standard into 17 files keeps that work contained. Changing one area means editing one file, instead of the same passage in four platform versions.

TPF is looking for international partners to contribute configurations, adaptations and clinical content for their own jurisdictions. That work is coordinated through the International Roundtable on Community Paramedicine (IRCP). [email protected]

Versions and contributions

Open, versioned, and open to correction

Version 3.0.0. Releases are published with notes on the Releases page. Updates are never automatic: replace the instructions and knowledge files as described in Updating an existing installation.

Outside scrutiny works

An outside contributor found a missing newborn (neonate) temperature threshold by comparing two copies of the vital sign table field by field. The contributor also worked out why cutting the text down by hand lost that value. Temperature was the only vital sign stated for a single age band, so its absence left no visible gap. That report is part of why 3.0.0 stopped relying on copies kept by hand.

Contribute
Clinical corrections or additions
Open a GitHub Issue with the change and your clinical reasoning. TPF reviews all contributions.
Contribute
Agency configurations
Share de-identified configurations for specific ePCR platforms so other agencies can adapt rather than start from scratch.
Contribute
International adaptations
Jurisdiction-specific standards, protocol references, scoring tool variations, and regulatory adaptations.
Contribute
Research and evaluation
Evaluations of tool output, narrative quality studies, and quality improvement (QI) integration findings, which inform future versions.

Opening a pull request? Edit the sources in src/, never the generated files at the repository root, which the build overwrites. Run python3 build.py && python3 build.py --verify before submitting.

Citation (APA 7)

If you use or adapt this work

PARAMEDIC-NARRATIVE by The Paramedic Foundation.

Research, policy work, and publications

Nudell, N. G. (2026). Paramedic-narrative-skill: AI-assisted PCR narrative documentation for paramedics and EMTs (Version 3.0.0) [Software]. The Paramedic Foundation. https://github.com/The-Paramedic-Foundation/paramedic-narrative-skill

Published by The Paramedic Foundation under CC BY 4.0 · GitHub repository · Latest release · ETHICS.md · [email protected]

Start with your next chart

Free, no registration, version 3.0.0. Published by The Paramedic Foundation under CC BY 4.0.

Download for Claude Choose another platform View on GitHub

Questions about agency configuration or governance consultation: [email protected]