Dockerpal

A terminal UI for exploring and cleaning up Docker: browse images, containers, networks and volumes, inspect them, and act on many at once without typing a single docker command.

It is built for the job docker image ls | grep ... | awk ... | xargs docker rmi usually gets thrown at: find the things you care about, tick off the ones you want, and do something to all of them.

• Every resource in one place — images, containers, networks and volumes, switched from a sidebar.
• Multi-select that sticks — mark rows with Space, and the selection follows the items through searching, filtering and refreshing.
• Search as you type — / filters the list on any column.
• Actions menu — a shows what the current screen can do, applied to the selection, with confirmation before anything is removed.
• Inspect and take away — full JSON for any resource, copied to the clipboard or written to a file, plus an approximate Dockerfile rebuilt from an image's history.
• Keyboard first — vi-style movement throughout, and every key is listed in the footer.

Install

pipx install dockerpal

Or into a virtualenv:

python -m venv .venv
source .venv/bin/activate
pip install dockerpal

Then run it:

dockerpal

Requirements: Python 3.11+, and a Docker daemon this user can reach — the usual /var/run/docker.sock, or wherever DOCKER_HOST points. If your user is not in the docker group, dockerpal starts up and tells you what went wrong rather than leaving you guessing. dockerpal --version prints the version.

Keys

Movement and selection work the same on every screen.

Containers add u start, x stop, t restart — each applied to the whole selection — and c to commit the container under the cursor as a new image.

In the details view, y copies and e exports; Esc goes back.

Actions menu

a lists what the current screen can do, with the shortcut for each. It acts on the selection, or on the row under the cursor when nothing is selected, and the title says which — Actions: 3 containers.

Force remove is menu-only and has no shortcut, so it cannot be hit by accident. Commit as image asks for a name (defaulting to <container>:latest) and works on the cursor row, since one image needs one name.

Search

/ filters the list as you type, matching any column case-insensitively — an image ID, part of a tag, a container status. Enter keeps the filter and returns to the table, Esc abandons it.

While rows are hidden the footer shows the query and the counts, as in Total: 7/63 Selected: 2 /nginx. Esc on the table then clears the filter; pressing it again quits. Selecting inside a filter and clearing it keeps your marks, so you can build a selection across several searches and then act on it once.

Inspecting, copying and exporting

Enter shows a resource's full JSON, syntax highlighted. There, y copies the selected text — or the whole document when nothing is selected — and e writes it to a file.

Images can also be exported straight from the actions menu, over a whole selection: Export JSON for the metadata and Export Dockerfile for an approximate Dockerfile. Files land in the current directory, named after the resource (nginx_1.27-alpine.json), and nothing is ever overwritten — an existing file gets a -1, -2 sibling.

Docker does not store the Dockerfile an image was built from, so the reconstruction comes from the commands recorded for each layer. Dockerpal undoes the formatting docker keeps them in, turning RUN |1 V=1 /bin/sh -c apk add git # buildkit back into RUN apk add git and EXPOSE map[80/tcp:{}] into EXPOSE 80/tcp. ADD/COPY contents are not in the image and BuildKit records less than the classic builder did, so treat the result as documentation rather than something to rebuild from.

Copying uses the platform's clipboard helper — wl-copy (Wayland), xclip or xsel (X11), pbcopy (macOS), clip.exe (WSL) — and also emits an OSC 52 escape sequence, which is what reaches the clipboard over ssh or inside tmux. With none of those installed only the escape sequence is sent, and the app says so.

Development

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt pytest pytest-asyncio -e .
pytest

The tests drive the real UI through Textual's test pilot against an in-memory stand-in for the Docker SDK (tests/fakes.py), so the suite needs no daemon and touches neither your containers nor your clipboard.

Status

Early days — it does what is described above and little more, and the interface may still change. Bug reports and ideas are welcome at the issue tracker.

License

MIT — see LICENSE.