A simple database for your own computer, made for Omarchy Linux.
Think of the old Microsoft Access, without the hard parts. Make a database, put a spreadsheet in it, look at your rows, print a tidy list. No SQL to learn, no server to set up, no account to make.
A jūbako is the stacked Japanese lunch box: one box, several tiers. Jubako is one file with several tables.
Status: usable, still young. Make a database, import a CSV or Excel file, browse and edit rows in a grid or one at a time in a form, export to CSV, Excel or PDF, and print a tidy report that fits the page. The command line and the MCP server for agents do the same jobs. See STATUS.md for exactly what works today.
- Make a database. It is one file on your computer. Copy it, back it up, email it. Or start from a template (Contacts, Inventory, Projects and tasks, Customer service, Home collection) that comes with its tables, saved views and reports, and a few example rows if you want them.
- Make a table. Name it and list its fields: what each is called, and whether it holds words, numbers, dates or yes/no. Then type in your rows.
- Import a spreadsheet. Drop a CSV or Excel file on the window. Jubako works out each column (words, whole numbers, numbers with decimals, dates, or yes/no), shows you its guesses, and lets you change them before anything is written. A workbook with several sheets can become one table per sheet.
- Look at your rows. A grid of everything in a table. Tap a row to pick it, double-tap a cell to change it.
- Type into a form. One record at a time: Previous, Next, Save, New row, Duplicate. Arrange it: field order, what each is called, which to leave out.
- Change your mind. Add a field, rename a field, delete a field, delete a table, or delete the whole database file. The destructive ones ask first. Mark a field "no repeats", like an email or a product code.
- Connect your tables. A lookup field points at a row of another table: an order's Item shows the product's name, and renaming the product shows everywhere. Turn a column of names into a lookup in one step; Jubako lists any name that matches nothing and can add it for you.
- See only some rows. "Show rows where Moved is not Yes" hides the rest in the grid and the form, and a printed report can use the same rule. Combine rules ("and" / "or"), compare numbers and dates ("is after 2024-01-01"), and tap a column heading to sort. Save what you keep using as a view, like "Still here", and it lives in the database file.
- Do something with those rows. Change a field in all of them, delete them, or see what they add up to (totals, averages, earliest and latest).
- Print a tidy report. Pick the columns, a title, the page size. Columns shrink and wrap so the whole table fits the page width. Sort it, group it with a heading per group, total the number columns, add your own header and footer line. See it before you print it, or save it as a PDF.
- Send it back out. Write any table to CSV, Excel (.xlsx) or PDF.
- Let an agent do it. The MCP server gives Claude Code (or any MCP client) the same jobs, so "turn this spreadsheet into a database" just works.
![]() |
![]() |
| Two rules and a sort, saved as a view: 32 of 242 orders | One record at a time, in a form you arrange |
![]() |
![]() |
| Grouped, totalled, and what you see is what prints | It follows your Omarchy theme, light or dark |
The screenshots use a made-up bike shop database: 242 orders, 86 customers, 28 products and 70 service jobs, imported from one Excel workbook.
Three pieces, one library underneath:
| Piece | What it is | Command |
|---|---|---|
| The app | A Qt Quick / QML desktop window. A small PySide6 host starts it and hands it a bridge to the library. Nothing about databases is written in QML. | jubako-app |
| Reports | One layout paints the on-screen preview, the PDF and the printer, so what you see is what you get. | in the app, jubako report, MCP |
| The command line | The same jobs from a terminal or a script. | jubako |
| The MCP server | The same jobs for agents, over stdio. | jubako-mcp |
The window follows your Omarchy theme. It reads the active theme's
colors.toml, so it is dark or light, and uses your accent colour, to match
the rest of the desktop. Change theme and the open window changes with it.
By default a new database is a single file in your Documents folder, ending in
.jubadb. It is an ordinary SQLite file, so nothing is locked away.
Jubako used to be called Omarchy-DB. Its .omadb files still open, and the
first open quietly brings them up to date.
If you already run a PostgreSQL or MySQL / MariaDB server, you can point Jubako at that instead when you make the database. Same app, same features; your data stays on your server.
| You pick | What it means | Needs |
|---|---|---|
| Just a file on this computer | The simple one. Nothing to install. | Nothing |
| A PostgreSQL server | A server you already run. | psycopg |
| A MySQL or MariaDB server | A server you already run. | PyMySQL |
Server passwords are never saved. The app remembers where the server is and asks for the password again next time.
Each release has packages for Arch (and Omarchy), Debian / Ubuntu, and Fedora. They pull in Python and Qt from your distro and put Jubako in your app launcher.
sudo pacman -U jubako-0.4.7-1-any.pkg.tar.zst # Arch, Omarchy
sudo apt install ./jubako_0.4.7-1_all.deb # Debian 13+, Ubuntu 26.04+
sudo dnf install ./jubako-0.4.7-1.noarch.rpm # Fedora 44+Excel files need openpyxl: apt and dnf install it along with Jubako, and on
Arch add python-openpyxl.
Or let autonomix-cli fetch the right package for your distro and keep it up to date:
autonomix-cli add https://github.com/ninepointlabs/jubakoEach release also has an AppImage, jubako-<version>-x86_64.AppImage, with its
own Python and Qt inside, so nothing else needs installing. It runs on any
x86_64 Linux from about 2021 on (glibc 2.34+: Ubuntu 22.04, Debian 12,
Fedora 39, openSUSE Leap 15.6, RHEL 9 and newer). Excel import and export,
PDF reports, printing and the PostgreSQL / MySQL backends all work from it.
chmod +x jubako-0.4.7-x86_64.AppImage
./jubako-0.4.7-x86_64.AppImage # the window
./jubako-0.4.7-x86_64.AppImage pets.jubadb # open a database straight away
./jubako-0.4.7-x86_64.AppImage cli tables pets.jubadb # the command line
./jubako-0.4.7-x86_64.AppImage mcp # the MCP server, for agentsTo get it in your app menu and dock, let Gear Lever or AppImageLauncher take it in: they use the menu entry and icons inside, and can update it in place when a new release comes out. Without one of those, the AppImage can do it itself:
./jubako-0.4.7-x86_64.AppImage --install-desktop-entry # menu entry, icons, .jubadb file type
./jubako-0.4.7-x86_64.AppImage --remove-desktop-entry # undo itThat writes to ~/.local/share (or $XDG_DATA_HOME) and nothing else, and
the entry points at wherever the AppImage is now, so run it again if you move
the file. On Omarchy the window follows your theme just as the package does;
elsewhere it is light or dark to match your desktop.
The AppImage mounts itself with FUSE. Its runtime carries its own FUSE
library, so the old advice to install libfuse2 does not apply; it only needs
the fusermount3 (or fusermount) helper that desktops ship. If it will not
start with a message about FUSE, install fuse3 (Ubuntu, Debian, Fedora,
openSUSE), or run it without FUSE:
./jubako-0.4.7-x86_64.AppImage --appimage-extract-and-runJubako needs Python 3.11 or newer. The window (and PDF reports) need Qt 6
and PySide6, which Omarchy already has (pyside6 and qt6-declarative).
# On Omarchy / Arch these are usually present already
sudo pacman -S --needed pyside6 qt6-declarative
git clone https://github.com/ninepointlabs/jubako.git ~/Projects/jubako
cd ~/Projects/jubako
uv venv --system-site-packages .venv # or: python -m venv --system-site-packages .venv
uv pip install -e '.[xlsx]' # or: .venv/bin/pip install -e '.[xlsx]'xlsx adds openpyxl for Excel files. Add a server backend only if you want one:
uv pip install -e '.[postgres]' # PostgreSQL
uv pip install -e '.[mysql]' # MySQL / MariaDBTo get it in your app launcher (links the commands into ~/.local/bin and
installs the desktop entry and icon under ~/.local/share):
./scripts/install-launcher.sh # --remove undoes itjubako-app # the home screen
jubako-app ~/Documents/pets.jubadb # open a database straight awayThe home screen has three big buttons: New database, Open a database, Import a spreadsheet, and the databases you opened lately. Drop a CSV or Excel file anywhere on the window and it becomes a database (or a table, if one is already open). Nothing is replaced without asking first.
Inside a database: New table… (under the list of tables) makes an empty table: give it a name and a line per field. Grid and Form show the same table two ways. Add row and Delete row do what they say. Show rows where… hides rows that do not match; Clear brings them back; Views… keeps a rule under a name and applies it again later. More… holds the rest: Add field…, the Fields list (rename or delete), delete the table, delete the whole database file. Print report opens the report designer with a live preview, Print… and Save PDF…. Export… writes the table as CSV or Excel. About (on the home screen, and at the bottom of the Database… menu) shows the version, license, Nine Point Labs and the support address, help@ninepointlabs.com.
jubako backends # what kinds of database you can make
jubako new "My Pets" ~/Documents/pets.jubadb # make one
jubako new-table ~/Documents/pets.jubadb Vets Name Phone "Last visit:date" # an empty table (unreleased: coming in the next release, not in 0.4.7)
jubako import ~/Documents/pets.jubadb pets.csv # put a spreadsheet in it
jubako import ~/Documents/pets.jubadb book.xlsx --all-sheets # one table per sheet
jubako tables ~/Documents/pets.jubadb # what's in there
jubako rows ~/Documents/pets.jubadb pets # look at the rows
jubako rows ~/Documents/pets.jubadb pets --filter moved is_not yes
jubako save-view ~/Documents/pets.jubadb pets "Still here" --filter moved is_not yes
jubako export ~/Documents/pets.jubadb pets out.csv # send it back out (--format csv|xlsx|pdf)
jubako report ~/Documents/pets.jubadb pets pets.pdf --landscape --title "All the pets"There is a tiny example to try in the source repository (it is not installed with the
package): data/examples/pets.csv.
Add this to your MCP client. For Claude Code:
claude mcp add jubako -- /home/you/Projects/jubako/.venv/bin/jubako-mcpWith the AppImage, point the client at the AppImage file with mcp as its
argument (or at a symlink to it named jubako-mcp):
claude mcp add jubako -- ~/Applications/jubako-0.4.7-x86_64.AppImage mcp{ "mcpServers": { "jubako": { "command": "/home/you/Applications/jubako-0.4.7-x86_64.AppImage", "args": ["mcp"] } } }Then ask it to do the work:
Take
~/Downloads/members.csvand make me a database in Documents.
The tools are list_backends, create_database, list_databases,
open_database, plan_import, import_spreadsheet, list_tables,
create_table, describe_table, list_rows, add_row, update_row, delete_row,
add_field, rename_field, delete_field, delete_table, delete_database,
list_views, get_view, save_view, delete_view,
create_form, get_form, create_report, list_reports, delete_report,
export_report and export_table. Run jubako-mcp --tools to print
their schemas.
When a spreadsheet comes in, each column gets the narrowest type that every filled-in cell fits. Blank cells never decide anything, and anything unclear stays as words, because words never lose data.
| The column holds | It becomes |
|---|---|
yes, no, true, 1, 0 |
Yes / No |
4, -3, 1,200 |
Whole number |
12.5, 6 |
Number with decimals |
2021-03-14, 14/03/2021 |
Date |
| anything else, or a mix | Words |
The guesses are shown to you before anything is written, and you can change them.
- Files stay where you said. Every path Jubako opens or writes is
checked against a list of approved folders — your home directory by default,
or whatever
JUBAKO_ROOTSsays. A path that climbs out with../, or a symlink that points outside, is refused. This matters most for the MCP server, where an agent supplies the paths. - Spreadsheet formulas are never run. In a CSV, a cell holding
=1+1is stored as the characters=1+1. In an Excel file, the value Excel last saved for the cell is used; the formula itself is never worked out, and macros are never touched. - Nothing is replaced quietly. Writing over a file or a table takes an
explicit
overwrite/--replace(the app asks you), and the result says what it replaced. - No passwords are kept. The list of recent databases remembers how to find a server, never how to log into one. Server passwords are passed in at the time you connect and are stripped out of anything printed or logged.
- Names are checked, not escaped. Table and field names must be plain identifiers; anything else is refused rather than quoted around.
Your own settings live in ~/.local/state/jubako/ (owner-only). Your
databases live wherever you put them.
uv venv --system-site-packages .venv
uv pip install -e '.[dev,postgres,mysql]'
.venv/bin/python -m pytestThe tests cover type guessing, path safety, the SQLite backend end to end, CSV and Excel import and export, reports (column fitting, pagination, PDF, the printer path), forms and reports kept in the database, the MCP server (including a real process over a pipe), and the app: the bridge, the rows model, editing, the import worker thread, the report preview, following a theme change, and loading the QML window offscreen with a table on screen.
A live PostgreSQL or MySQL server is not needed for the test suite. To check one of those for real:
JUBAKO_PASSWORD=secret .venv/bin/python scripts/smoke_remote.py postgres \
--host localhost --database jubako_smoke --user postgresscripts/build-packages.sh --test # Arch, .deb and .rpm into dist/, then install-test eachIt builds in Docker containers from the committed tree (--ref v0.4.7 for a
tag), then installs each package in a fresh Debian, Ubuntu, Fedora and Arch
container and checks the command line, the MCP server and the window. When
pkgver changes, release the tag, then set sha256sums in
packaging/arch/PKGBUILD to the tag tarball's checksum.
scripts/build-appimage.sh --test # the AppImage into dist/, then packaging/appimage/smoke.sh on itThe AppImage is built on Ubuntu 22.04 (in Docker here, on GitHub Actions for
a release) from pinned, checksummed downloads: python-appimage's Python 3.12,
the PySide6 wheels and the rest of packaging/appimage/requirements.txt, and
appimagetool. Publishing a release runs .github/workflows/appimage.yml,
which builds it, checks it, and attaches the AppImage, its .zsync and
.sha256 to the release and adds it to SHA256SUMS. It only ever adds
files. For a release that already exists, run the workflow by hand with its
tag.
src/jubako/ the core: backends, import, export, reports, printing, paths, CLI
src/jubako_app/ the desktop app: PySide6 host, bridge, report bridge, theme, and qml/
src/jubako_mcp/ the MCP server agents attach to
archive/gtk-prototype/ the first (GTK) window, kept for reference only
data/examples/ a small sample spreadsheet
tests/ the test suite
packaging/ the desktop entry, icon, PKGBUILD, rpm spec, and the .deb/.rpm staging script
scripts/ launcher install, package builds, and the server-backend smoke test
MIT. See LICENSE.




