Files
temetro/CHANGELOG.md
T
Khalid Abdi b29fdff1cb feat: ambient AI visit scribe (record/paste → reviewed SOAP note)
Record a clinician↔patient visit (or paste a transcript) on the patient
sheet; the backend transcribes it (OpenAI Whisper / Gemini), de-identifies
the transcript + context through Veil, and drafts a structured SOAP note
the clinician reviews and edits before saving — the same write-approval
gate as the chat agent.

Backend: POST /api/scribe/{transcribe,draft,save} (routes/scribe.ts,
services/ai/transcribe.ts), veil.redactText() free-text redactor,
appendEncounter service, audio MIME types on attachments. Gated by
patient:write + the clinic AI policy (reception/disabled-AI excluded).
Frontend: ScribeDialog + lib/scribe.ts, "Record visit" on the patient
detail, gated by clinical access + AI availability. New `scribe` locale
namespace across all five languages. Bumps to v0.4.0.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-03 18:33:30 +03:00

164 lines
8.6 KiB
Markdown

# Changelog
All notable changes to temetro are recorded here. The format is loosely based on
[Keep a Changelog](https://keepachangelog.com/), and the project follows
[Semantic Versioning](https://semver.org/). See [RELEASING.md](./RELEASING.md)
for how releases are cut and published.
## [Unreleased]
## [0.4.0] — 2026-07-03
### Added
- **Ambient AI visit scribe** — a **Record visit** action on the patient sheet turns a
clinician↔patient conversation into a draft **SOAP** encounter note. Record with the
microphone (`MediaRecorder`, stored as an auditable patient attachment) or paste a
transcript; the backend transcribes via the user's **OpenAI (Whisper)** or **Gemini**
key, de-identifies the transcript + patient context through **Veil**, and the model
drafts a structured note that the clinician **reviews and edits before saving** — the
same write-approval gate as the chat agent. New `POST /api/scribe/{transcribe,draft,save}`
(`backend/src/routes/scribe.ts`, `services/ai/transcribe.ts`), a `veil.redactText()`
free-text redactor, and an `appendEncounter` service that adds one note without touching
the rest of the record. Gated by `patient:write` + the clinic AI policy (reception and
disabled-AI accounts don't see it). New `scribe` locale namespace across all five
languages. Drafting also works with local Ollama from a pasted transcript.
## [0.3.0] — 2026-07-02
### Added
- **Three new interface languages** — Somali (`so`), Arabic (`ar`) and German
(`de`) join English and French, selectable in Settings → Profile → Language.
Each locale carries a full translation of the ~1,660 UI strings, with native
names shown in the selector (Soomaali, العربية, Deutsch).
- **Right-to-left (RTL) support** — selecting Arabic sets `dir="rtl"` on the
document (applied before first paint via an inline script, so no flash), flips
physical spacing/alignment to logical CSS utilities, mirrors directional icons,
and loads an Arabic-capable typeface (IBM Plex Sans Arabic) appended to the
font stack for per-character fallback.
- **Language roams across devices** — the chosen language is persisted to the
per-user `user_settings` preferences and re-applied on sign-in, with
localStorage remaining the offline source of truth.
- **`frontend/scripts/check-locales.mjs`** (+ `npm run check-locales`) — a parity
check that fails on missing/extra keys or `{{placeholder}}` mismatches across
locales and warns when Arabic count-keys lack the full CLDR plural forms.
## [0.2.5] — 2026-07-01
### Fixed
- **Multi-arch Docker images** — the `release` workflow now builds and publishes
`khalidxv/temetro-backend` and `khalidxv/temetro-frontend` for both
`linux/amd64` and `linux/arm64`. Previously the images were amd64-only, so
`docker compose pull` on Apple Silicon failed with *no matching manifest for
linux/arm64/v8* and fell back to building from source.
### Changed
- **Language switcher** in Settings → Profile is now a **select** that asks for
confirmation before switching the interface language, instead of applying the
change instantly on a button tap.
## [0.2.4] — 2026-06-29
### Added
- **Pagination** on the Activity and Invoices pages (10 per page), matching the
Patients page, via a shared `ListPagination` component.
- **French (Français)** interface language, with a language switcher in
Settings → Profile. The choice persists on the device.
- **"Check for updates"** button in Settings → About & updates that forces a
fresh check.
### Changed
- **Update detection** now reads the latest version from **Docker Hub** image
tags (the channel clinics actually pull), falling back to the GitHub release
if Docker Hub is unreachable. This fixes "About & updates" showing *Up to
date* when a newer image was already published.
- **docker compose** host ports are now configurable (`BACKEND_PORT`,
`FRONTEND_PORT`, `ADMINER_PORT`, alongside `POSTGRES_PORT`) so a port clash on
`docker compose up -d` can be resolved from `.env` without editing the file.
## [0.2.3] — 2026-06-29
### Fixed
- **AI chat patient record cards** now render on Google Gemini for name
lookups. "Show me <name>'s medical record" relied on the model chaining
`searchPatients``getPatient`, but Gemini often called `searchPatients`
and then emitted only a canned closing line ("Here's the record.") without
the second tool call — so no card was ever drawn. `searchPatients` now
displays the record card directly when exactly one patient matches, so the
flow no longer depends on a follow-up tool call. (The previous Gemini fix in
0.2.2 only covered the empty-schema list tools.)
## [0.2.2] — 2026-06-28
### Fixed
- **AI chat record cards** now render on Google Gemini. The card-emitting
chat tools (`listAppointments`, `listTasks`, `listPrescriptions`,
`getClinicInfo`, `getAnalytics`, `listInventory`) used an empty parameter
schema; Gemini can't emit a function call for a schema with no properties,
so it printed the call as `tool_code` text instead of invoking the tool —
leaving replies as plain text (e.g. "Show today's schedule" leaked a raw
`<tool_code>` block) with no cards. The tools now share a non-empty schema
so Gemini calls them; other providers are unaffected.
## [0.2.1] — 2026-06-27
### Fixed
- **Patients pagination** controls now render as proper buttons — the prev/next
and page-number controls were unstyled and wrapping (the COSS `PaginationLink`
drops its button styling when given a `render` prop).
- **Patient detail sheet header** reflowed: actions (Download summary / Transfer
/ Edit / Delete) moved to their own wrapping row so the patient name is no
longer truncated.
- **Messages thread** now shows **sender and recipient avatars** alongside the
chat bubbles.
- **Release notes** — the `release` workflow now publishes the matching
`CHANGELOG.md` section as the GitHub Release body (instead of only the
auto-generated "Full Changelog" link).
## [0.2.0] — 2026-06-27
### Added
- **Patients table pagination.** The Patients list now paginates at 10 rows per
page (COSS `Pagination`), so large clinics no longer scroll endlessly.
- **Per-patient record history.** The patient detail sheet shows an audit
timeline of every add/change on that chart. `GET /api/activity/patient/:fileNumber`.
- **Patient summary PDF.** A **Download summary** action on the patient sheet
produces a clean, printable one-page clinical summary (browser "Save as PDF").
- **AI setup notice.** A single, dismissible heads-up appears above the chat
input on a fresh chat when no AI provider (API key or local Ollama) is
configured; it clears itself once you send a message.
- **Version & update awareness.** `GET /api/version` reports the running version
and checks GitHub Releases for a newer one; Settings → **About & updates** shows
the current/latest version, and an optional, dismissible banner appears when an
update is available.
- **LAN access.** The frontend now resolves the backend URL from the host the
browser is using, so other departments can reach temetro at
`http://<server-LAN-IP>:3000` with no rebuild. A Settings panel surfaces the
shareable network address. `GET /api/network` reports detected LAN addresses.
- **Prebuilt Docker images** published to Docker Hub (`khalidxv/temetro-backend`,
`khalidxv/temetro-frontend`) via a tag-triggered GitHub Actions release workflow.
- **Voice dictation** on the AI chat input (Web Speech API), with graceful
fallback where the browser doesn't support it.
### Changed
- **Messages thread** rebuilt on the shadcn `Message` / `Bubble` / `Attachment`
components (COSS colour tokens preserved) for a cleaner conversation surface.
- **Patient status badges** now use semantic colours (active → success,
inpatient → info) instead of a flat secondary badge.
- `docker-compose.yml` references the published images (with a build fallback) and
no longer bakes a fixed API URL into the frontend.
### Fixed
- **AI import approval card.** `previewImport` now declares a concrete record
schema so Google Gemini emits a real tool call (and the approval card renders)
instead of dumping a `tool_code`/JSON wall as text. The system prompt also
forbids printing tool calls and re-listing fields.
- **Settings network address** no longer shows a bogus container IP / "Error"
under Docker — the `/api/network` endpoint now skips container-internal
interfaces, and the panel falls back to the helpful LAN hint.
## [0.1.0] — 2026-06-26
Initial baseline: clinician AI-chat UI wired to the TypeScript/Express/Postgres
API (Better Auth, multi-tenant clinics, org-scoped patient records), the patient
wallet encrypted-share flow, and Dockerised local run.