WEBVTT

00:00:00.000 --> 00:00:02.767
QA Scribe is a local-first testing notepad

00:00:02.767 --> 00:00:05.533
for people who turn exploratory testing into

00:00:05.533 --> 00:00:08.300
reusable Testware. It captures the raw work

00:00:08.300 --> 00:00:11.067
of a testing Session, then helps shape

00:00:11.067 --> 00:00:13.833
that material into Findings and Drafts. The

00:00:13.833 --> 00:00:16.600
important boundary is local control: storage is

00:00:16.600 --> 00:00:19.367
local SQLite, attachments are local files, and

00:00:19.367 --> 00:00:22.133
AI runs through local command-line tools. The

00:00:22.133 --> 00:00:24.900
user flow starts with a tester writing

00:00:24.900 --> 00:00:27.667
notes during a focused Session. Those notes

00:00:27.667 --> 00:00:30.433
can include context, observations, logs, API responses

00:00:30.433 --> 00:00:33.200
screenshots, and follow-up thoughts. When the tester

00:00:33.200 --> 00:00:35.967
is ready, the same material can become

00:00:35.967 --> 00:00:38.734
editable Testware or a structured Finding. The

00:00:38.734 --> 00:00:41.500
app keeps those records linked back to

00:00:41.500 --> 00:00:44.267
the testing evidence. The runtime has three

00:00:44.267 --> 00:00:47.034
layers. The React renderer owns the desktop

00:00:47.034 --> 00:00:49.800
interaction model. The Tauri bridge exposes typed

00:00:49.800 --> 00:00:52.567
commands to the renderer. The Rust core

00:00:52.567 --> 00:00:55.334
owns storage, validation, managed files, generation prompts

00:00:55.334 --> 00:00:58.100
and response persistence. SQLite and the attachment

00:00:58.100 --> 00:01:00.867
directory sit behind the core service layer

00:01:00.867 --> 00:01:03.634
Provider CLIs sit outside the app and

00:01:03.634 --> 00:01:06.400
are only called after explicit user action

00:01:06.400 --> 00:01:09.167
Session capture moves through a narrow path

00:01:09.167 --> 00:01:11.934
The editor calls a frontend action factory

00:01:11.934 --> 00:01:14.700
The action calls a typed Tauri command

00:01:14.700 --> 00:01:17.467
The command locks the app state and

00:01:17.467 --> 00:01:20.234
calls SessionService. SessionService validates ownership and writes

00:01:20.234 --> 00:01:23.000
the SQLite rows. That path keeps the

00:01:23.000 --> 00:01:25.767
UI simple and keeps data rules in

00:01:25.767 --> 00:01:28.534
Rust. The domain model is intentionally small

00:01:28.534 --> 00:01:31.300
A Session scopes the work. An Entry

00:01:31.300 --> 00:01:34.067
stores raw testing material. A Finding records

00:01:34.067 --> 00:01:36.834
a structured conclusion. A Draft stores editable

00:01:36.834 --> 00:01:39.600
Testware. Evidence connects conclusions back to Entries

00:01:39.600 --> 00:01:42.367
and managed attachments. An AI Run records

00:01:42.367 --> 00:01:45.134
each generation request and its outcome. Attachments

00:01:45.134 --> 00:01:47.900
are managed instead of being loose file

00:01:47.900 --> 00:01:50.667
references. When an image is pasted or

00:01:50.667 --> 00:01:53.434
uploaded, the backend sanitizes the path, writes

00:01:53.434 --> 00:01:56.201
it under the Session attachment folder, hashes

00:01:56.201 --> 00:01:58.967
the bytes, and stores metadata in SQLite

00:01:58.967 --> 00:02:01.734
Preview reads verify the hash before returning

00:02:01.734 --> 00:02:04.501
a data URL. Rich editor images use

00:02:04.501 --> 00:02:07.267
a QA Scribe attachment protocol instead of

00:02:07.267 --> 00:02:10.034
arbitrary local file paths. AI generation is

00:02:10.034 --> 00:02:12.801
explicit. The frontend opens a preflight step

00:02:12.801 --> 00:02:15.567
where the user chooses an action, provider

00:02:15.567 --> 00:02:18.334
model, and reasoning effort. The backend prepares

00:02:18.334 --> 00:02:21.101
a Generation Context and a running AI

00:02:21.101 --> 00:02:23.867
Run. Then a local Claude, Codex, or

00:02:23.867 --> 00:02:26.634
Copilot CLI receives the prompt on standard

00:02:26.634 --> 00:02:29.401
input. Streaming events update the UI until

00:02:29.401 --> 00:02:32.167
the final Draft, Finding, or summary is

00:02:32.167 --> 00:02:34.934
persisted. Provider readiness is a separate system

00:02:34.934 --> 00:02:37.701
QA Scribe checks whether each local CLI

00:02:37.701 --> 00:02:40.467
is installed and authenticated. It detects model

00:02:40.467 --> 00:02:43.234
choices where the CLI exposes them, caches

00:02:43.234 --> 00:02:46.001
readiness briefly, and never stores provider API

00:02:46.001 --> 00:02:48.767
keys. This keeps authentication with the provider

00:02:48.767 --> 00:02:51.534
tool that already owns it. On the

00:02:51.534 --> 00:02:54.301
frontend, one controller coordinates most app state

00:02:54.301 --> 00:02:57.067
useAppController owns the active Session, note body

00:02:57.067 --> 00:02:59.834
Drafts, Findings, provider state, generation jobs, busy

00:02:59.834 --> 00:03:02.601
flags, and notices. AppShell lays out the

00:03:02.601 --> 00:03:05.367
desktop frame. Views render Notes, Testware, Findings

00:03:05.367 --> 00:03:08.134
and Settings. The editor and workflow modules

00:03:08.134 --> 00:03:10.901
handle rich text, preflight, copy, and deletion

00:03:10.901 --> 00:03:13.668
flows. The codebase is compact but broad

00:03:13.668 --> 00:03:16.434
There are eighty-two Rust files, fifty-six TypeScript

00:03:16.434 --> 00:03:19.201
and TSX files, and a release pipeline

00:03:19.201 --> 00:03:21.968
that builds desktop artifacts and Linux package

00:03:21.968 --> 00:03:24.734
repository metadata. Recent churn concentrates around provider

00:03:24.734 --> 00:03:27.501
process execution, generated bindings, frontend controller tests

00:03:27.501 --> 00:03:30.268
and release automation. Those are the areas

00:03:30.268 --> 00:03:33.034
to test carefully after changes. You can

00:03:33.034 --> 00:03:35.801
remember the repository by its seams. Change

00:03:35.801 --> 00:03:38.568
domain rules in qa-scribe core. Change desktop

00:03:38.568 --> 00:03:41.334
commands in src-tauri. Change user workflows in

00:03:41.334 --> 00:03:44.101
frontend source. Change packaging in scripts and

00:03:44.101 --> 00:03:46.868
GitHub workflows. The design goal is steady

00:03:46.868 --> 00:03:49.634
capture testing evidence locally, keep it traceable

00:03:49.634 --> 00:03:52.401
and turn it into useful Testware without

00:03:52.401 --> 00:03:55.168
taking control away from the tester. For

00:03:55.168 --> 00:03:57.934
a new contributor, the safest starting point

00:03:57.934 --> 00:04:00.701
is the workflow seam closest to the

00:04:00.701 --> 00:04:03.468
change. If the behavior is about saved

00:04:03.468 --> 00:04:06.234
data, start in the Rust service tests

00:04:06.234 --> 00:04:09.001
If it is about how a tester

00:04:09.001 --> 00:04:11.768
moves through the app, start in the

00:04:11.768 --> 00:04:14.534
React action factories and view tests. If

00:04:14.534 --> 00:04:17.301
it is about generation reliability, start with

00:04:17.301 --> 00:04:20.068
the job runner, stream executor, and provider

00:04:20.068 --> 00:04:22.835
detection tests. That split makes the repository

00:04:22.835 --> 00:04:25.601
easier to reason about than its desktop

00:04:25.601 --> 00:04:28.368
shape first suggests. That is the tour

00:04:28.368 --> 00:04:31.135
a small desktop app, a clear core

00:04:31.135 --> 00:04:33.901
service boundary, and a testing workflow that

00:04:33.901 --> 00:04:36.668
preserves the tester's evidence. The useful mental

00:04:36.668 --> 00:04:39.435
model is capture first, structure second, and

00:04:39.435 --> 00:04:42.201
generate only when the user asks for

00:04:42.201 --> 00:04:42.597
it.