Feature Requests

Currently, to generate documents or use Ask AI with the patient file, it is mandatory to create a session and select active documents. Michalina suggests having an Ask AI window directly integrated into the patient record screen, with all patient information already available and accessible. This would make it possible to generate documents, write correspondence or create a global psychotherapy report without going through a session, which would greatly facilitate administrative work and communication.

I've been thinking about a potential development idea for using AI within general practice that deliberately stays away from clinical decision making. Rather than asking AI to diagnose or recommend treatment, the concept would be to create a practice-specific knowledge base containing our local protocols, referral pathways, patient information leaflets, formulary preferences, standard advice, administrative processes and communication templates. The clinician would still make all of the clinical decisions as they do now. AI's role would simply be to: search the local knowledge base, identify the relevant practice-specific protocol or process, and generate a personalised patient email, text message or consultation note based on the clinician's chosen management plan. For example: recurrent impetigo → clinician decides to prescribe clarithromycin and request a swab → AI produces our practice email explaining how to collect the swab, hygiene advice, safety netting and documentation for the notes. CDC referral → AI automatically includes our local Plymouth referral process, school forms, links and advice. Blood tests → AI inserts our surgery-specific booking instructions rather than generic NHS advice. Fit notes → AI generates the appropriate wording based on our local process. Referrals → AI uses local referral criteria and attaches the correct patient information. Verrucas- general advice The important distinction is that AI isn't deciding what to do. The clinician decides the management plan; AI simply applies local policies and produces consistent, high-quality patient communications and documentation. From an IG perspective, this feels much lower risk because the model isn't being trained on patient data or making autonomous clinical decisions. It's effectively acting as an intelligent document and protocol assistant. I think there is real potential to save clinicians several minutes per consultation while also improving consistency, reducing variation and ensuring patients receive the correct local information every time. I'd be really interested in whether this fits with the sort of AI development work NHS Devon is exploring.

When you copy from ask Heidi it would be better if it wasn't copied in markdown format

User wants to disable the Enter key from submitting queries in Ask Heidi when typing multi-line content like op reports and information sheets. Currently, pressing Enter to move to a new line triggers query submission, which is frustrating for users with 20+ years of typing surgical notes who rely on Enter for paragraph breaks. This affects workflow when creating multi-line documents outside of live consulting.

I use “Ask Heidi” to make my radio prescriptions based on the elements of the transcript and the context. I must always specify to him: “X-ray prescription with clinical elements” for the prescription to contain clinical elements. I would like to give him a permanent rule: “when I write “radio”, it means that I am asking you for an X-ray order specifying the relevant clinical elements, based on the transcript and the context.” This would greatly simplify requests to Heidi IA during consultation.

It would be useful to be able to highlight a line or number of lines within the note editor, press a keyboard shortcut to activate a dictation/typing window and then dictate/write changes for an AI model to make rather than directly making the changes. This would be similar to the inline editing within Cursor or VS Code. Ideally, it would use a smaller/faster model to allow for rapid changes/prototyping.

This was available on my previous provider, and it's a feature I miss.

Would it be possible to have a system to feed AI with various examples for a given model? This would make it possible to enrich the model and to specify situations where the model will have to react in a particular way.

Greetings. Some feedback on user experience. I am finding that I push the transcribe button, then start typing the patient's name, but after a few seconds realise that the pop-up for the "ask Heidi anything" has popped up and a portion of the name has been written into the "ask Heidi anything" pop-up. Is it possible to delay the pop-up by eg 10 seconds, to allow for entering patient name? I realise that an easy work-around is to do the patient's name first, but I want to start the transcription so as not to miss the early part of the consultation. (PS this is mostly an issue when things are very busy) Alternatively, is it possible to display the pop-up, but not move the cursor to the pop-up?