1
00:00:00,000 --> 00:00:03,133
Data modeling for Snowflake should not require a

2
00:00:03,133 --> 00:00:06,267
multi-thousand-dollar enterprise license. I wanted a practical tool

3
00:00:06,267 --> 00:00:09,400
that could reverse engineer a real database, give

4
00:00:09,400 --> 00:00:12,534
me an editable logical and physical model, and

5
00:00:12,534 --> 00:00:15,667
generate Snowflake-ready DDL. So I assigned the mission

6
00:00:15,667 --> 00:00:18,800
to an autonomous AI employee and let it

7
00:00:18,800 --> 00:00:21,479
build on proven open-source foundations.

8
00:00:21,500 --> 00:00:24,784
This is the packaged Apple Silicon application working

9
00:00:24,784 --> 00:00:28,068
with a live Chinook schema in Snowflake. The

10
00:00:28,068 --> 00:00:31,352
app discovered eleven tables, sixty-four columns, primary keys,

11
00:00:31,352 --> 00:00:34,636
and eleven foreign-key relationships. It converted that metadata

12
00:00:34,636 --> 00:00:37,920
into an editable diagram and laid the tables

13
00:00:37,920 --> 00:00:41,204
out automatically. This is not a mock-up; it

14
00:00:41,204 --> 00:00:45,208
is the application running on the Mac build.

15
00:00:45,208 --> 00:00:48,089
The experience starts as a desktop data-modeling tool,

16
00:00:48,089 --> 00:00:50,970
not a browser tab that must stay connected

17
00:00:50,970 --> 00:00:53,852
to a development server. You can begin with

18
00:00:53,852 --> 00:00:56,733
a blank Snowflake model, create and edit tables

19
00:00:56,733 --> 00:00:59,614
and relationships, arrange the canvas, and work with

20
00:00:59,614 --> 00:01:02,495
the familiar drawDB interface inside a native Electron

21
00:01:02,495 --> 00:01:03,575
shell.

22
00:01:03,608 --> 00:01:06,779
Reverse Engineer Snowflake is a first-class desktop action.

23
00:01:06,779 --> 00:01:09,950
The application can use an existing Snowflake CLI

24
00:01:09,950 --> 00:01:13,122
profile or a deliberately bounded manual connection. Credentials

25
00:01:13,122 --> 00:01:16,293
and sessions stay in Electron's main process and

26
00:01:16,293 --> 00:01:19,464
are never written into the ERD project file.

27
00:01:19,464 --> 00:01:22,635
Select a database, schema, and tables, then import

28
00:01:22,635 --> 00:01:26,526
the result as a model you can change.

29
00:01:26,526 --> 00:01:30,062
The same model generates deterministic Snowflake DDL. Database

30
00:01:30,062 --> 00:01:33,599
and schema creation, table definitions, data types, comments,

31
00:01:33,599 --> 00:01:37,135
primary keys, and foreign keys come from the

32
00:01:37,135 --> 00:01:40,671
canonical model rather than from a one-off text

33
00:01:40,671 --> 00:01:44,208
prompt. Snowflake constraints are labeled informational and not

34
00:01:44,208 --> 00:01:47,744
enforced, matching the platform's behavior instead of pretending

35
00:01:47,744 --> 00:01:51,116
they operate like transactional database constraints.

36
00:01:51,126 --> 00:01:54,791
Projects save as native, credential-free ERD files and

37
00:01:54,791 --> 00:01:58,456
reopen in a fresh desktop process. The model,

38
00:01:58,456 --> 00:02:02,122
diagram layout, comments, constraints, and relationships survive the

39
00:02:02,122 --> 00:02:05,787
round trip. That gives the tool a real

40
00:02:05,787 --> 00:02:09,452
document workflow: reverse engineer, refine the design, save

41
00:02:09,452 --> 00:02:13,117
it, reopen it later, and regenerate the DDL

42
00:02:13,117 --> 00:02:15,670
without reconnecting to Snowflake.

43
00:02:15,693 --> 00:02:18,706
The product story matters as much as the

44
00:02:18,706 --> 00:02:21,720
feature list. The AI employee first proved the

45
00:02:21,720 --> 00:02:24,733
pipeline by reverse engineering SQLite and forward engineering

46
00:02:24,733 --> 00:02:27,746
the structures into Snowflake. I then assigned the

47
00:02:27,746 --> 00:02:30,760
next mission: turn the open-source web application into

48
00:02:30,760 --> 00:02:33,773
an installable Snowflake modeling app. It worked through

49
00:02:33,773 --> 00:02:36,787
the backlog, fixed orchestration and product failures at

50
00:02:36,787 --> 00:02:40,143
their source, and delivered this Mac build.

51
00:02:40,144 --> 00:02:43,603
The Apple Silicon package has passed one hundred

52
00:02:43,603 --> 00:02:47,062
fifteen automated checks plus a live acceptance run:

53
00:02:47,062 --> 00:02:50,522
connect, discover, reverse engineer, save, reopen, and regenerate

54
00:02:50,522 --> 00:02:53,981
forward DDL. The current build is unsigned and

55
00:02:53,981 --> 00:02:57,441
intended for private preview. Linux, Windows, signing, and

56
00:02:57,441 --> 00:03:00,900
notarization remain future release work. To follow the

57
00:03:00,900 --> 00:03:04,359
project or discuss a Snowflake modeling use case,

58
00:03:04,359 --> 00:03:07,242
visit Lee Harrington's project hub.