v2.1.4 — Streaming Notes, Smarter Prompts, and a Quieter Patient Loop
There's a stretch of work we just shipped that, taken together, says something about where Aisel is going. Not a single headline feature — a shift in how the platform behaves around the clinician.
The note now writes itself in front of you. The prompts know what kind of conversation they're shaping. The patient can answer back, and you can see it. And a handful of small frictions — the ones you stopped noticing because you'd worked around them — are gone.
Here's what changed.
A new note editor that streams
The biggest piece of this release is the v2 AI note editor.
Notes now stream in as they're generated. You watch the structure form, section by section, rather than waiting for a finished block to appear. You can revise a section, regenerate it, or let it autosave as you go. Every note keeps a version history, so stepping back to an earlier draft is one click — not a guess at what you used to have.
It's built on a new editing foundation (Lexical), with a streaming pipeline behind it (server-sent events, per-section dispatch). What that means in practice: the editor is faster, more responsive, and finally feels like a place to work on a note rather than a place to receive one.
The v2 editor is rolling out behind a feature flag. If you don't see it yet, you will.
Prompts that know what they're for
Until now, the same underlying prompt shaped every kind of conversation Aisel processed — a Typeform intake, an ElevenLabs voice conversation, a live consult. They're not the same kind of input, and they shouldn't be treated the same.
This release introduces scoped prompts. Each conversation type can now have its own tuned prompt, with admin controls to manage the overrides. Prompt versioning and ranking were also fixed under the hood — so when there are multiple candidates, the right one wins, every time.
You won't see this directly. You'll feel it in the output.
Clinicians can sign themselves up
Self sign-up for clinicians is live. No invitation flow, no manual provisioning step. A clinician who wants to try Aisel can start.
This is the smaller of the changes in this release, but it changes the shape of how new clinicians arrive on the platform.
Patient feedback, surfaced to the clinician
Patients can now leave structured feedback after a clinical conversation — and the clinician sees it directly in the platform.
This closes a loop that, until now, was open. The patient experiences the conversation; the clinician experiences the documentation; the two halves don't usually meet. Now they do. If a patient flags that something felt off — or right — the clinician knows.
It's a small surface. We expect it to do more over time.
Corti: ambient recording, properly tuned
For clinicians using Corti for ambient capture, a cluster of fixes that add up to a meaningful quality difference:
- Audio now streams in 250ms chunks, the cadence Corti recommends — smoother, less choppy transcription.
- Swiss German is correctly recognised as
gsw-CH. It was being mishandled. - Input language and output language are now separately configurable.
- Ambient single-channel recording now uses the right participant role, so transcription doesn't get confused about who's speaking.
The Corti integration was usable before. It's tighter now.
Patient and session polish
The small things you stopped noticing:
- Note templates are now sorted alphabetically, so finding the one you want is finally predictable.
- Patient quick-select sticks when you create a note — no more re-prompt for the patient you already chose.
- Draft notes with a patient open in the patient's information page, where they belong.
- Copy/paste preserves formatting. It used to flatten.
- Revert to draft is back for completed notes — useful when you want to revise something already signed.
- Patient session reassignment has been fixed for the edge cases that were quietly failing.
Typeform PDFs, translated
Intake PDFs from Typeform are now translated into the clinician's working language, with an updated layout that's easier to read at a glance.
A note on what we're trying to do
The throughline across this release isn't "more features." It's that the platform is getting better at staying out of the clinician's way while doing more underneath.
The editor streams so you don't wait. The prompts adapt so you don't have to think about them. The patient can answer so the loop closes on its own. The fixes around copy-paste and quick-select and template sorting are there because every one of them was a tiny tax on your attention.
That's the work. We'll keep at it.