NKUhan123/dsh-jupyterlab

A JupyterLab-style notebook for DeepSeek Harness. A real, persistent ipykernel the agent can read, edit and run — with a front end that behaves like JupyterLab: code on the left, figures on the right, Jupyter keyboard shortcuts, and Markdown cells.

This is a modified fork of @beihaizb/dsh-notebook by beihzb (MIT). The kernel sidecar, the nb_* agent tools, .ipynb save/load and the cell UI are upstream work. This package keeps all of it and adds the features below, as a single self-contained plugin (the upstream sources are vendored, so dsh plugin add installs everything).

What this fork adds

1. Split view (code | output). Every code cell renders as two columns: the editor on the left, its outputs — figures, stdout, DataFrames, tracebacks — on the right. A draggable divider resizes the columns and the ratio is remembered. The toolbar button ⇔ Split / ⇕ Inline returns to the classic stacked layout.
2. Outputs stay in sight. The output pane is bottom-aligned and sticks to the viewport while you scroll a long cell, so a figure you just rendered is already on screen instead of being stranded at the top of the cell.
3. Jupyter keyboard shortcuts. Press Esc to leave editing (command mode), then a / b insert a cell above / below, dd deletes, Enter edits, ↑ ↓ (or k j) move, m / y switch the cell to Markdown / Code, and Ctrl+Enter / Shift+Enter / Alt+Enter run. While editing, Tab accepts the completion popup (and still indents when none is open). ? opens the shortcut list.
4. Cell type and one-click delete. A ✂ scissors in the gutter deletes the cell, and a py / md chip shows and switches the cell type. Markdown cells render ## headings, are edited by double-click, and can be folded like code cells.
5. + MD in the toolbar inserts a Markdown cell (the existing + Cell now selects and focuses what it inserts).

Everything else — persistent kernel, agent tools, .ipynb round-trip, kernel picker, tqdm progress bars, clickable tracebacks, long-output folding, hybrid-CPU hint, EN/中文 UI — is upstream and unchanged.

Install

From the GitHub repository (this is what the plugin market uses):

dsh plugin --profile desktop add github:NKUHan123/dsh-jupyterlab

Or from the prebuilt tarball attached to the latest GitHub Release (no build step):

dsh plugin --profile desktop add file:<path-to>/dsh-jupyterlab.tgz  # prebuilt tarball from the latest GitHub Release

Then restart DeepSeek Harness. A Notebook tab appears in the session.

The package ships its compiled lib/ and the Python driver, so there is nothing to build at install time (no prepare / install scripts — pnpm has no build to approve).

Don't install this next to @beihaizb/dsh-notebook. Both register the same nb_* tools; this fork is a superset, so remove the original first:

dsh plugin --profile desktop remove @beihaizb/dsh-notebook

The Python environment you pick as kernel needs ipykernel, jupyter_client and nbformat.

Keyboard

Mode	Key	Action
command	Enter	edit the selected cell
command	Esc	back to command mode
command	a / b	insert a cell above / below
command	d d	delete the selected cell
command	↑ ↓ / k j	select previous / next cell
command	m / y	switch to Markdown / Code
command	Ctrl+Enter	run, stay on the cell
command	Shift+Enter	run, move to the next cell
command	Alt+Enter	run, insert a cell below
command	?	shortcut list
editing	Ctrl+Enter / Shift+Enter / Alt+Enter	same run modes
editing	Tab	accept completion (indent when nothing is open)
editing	Esc	leave editing

Tools (unchanged from upstream)

nb_new / nb_open / nb_save / nb_get / nb_context / nb_list / nb_list_vars / nb_inspect_object / nb_add_cell / nb_delete_cell / nb_move_cell / nb_edit_cell / nb_edit_and_run_cell / nb_cell_history / nb_revert_cell / nb_error_context / nb_run_cell / nb_run_all / nb_apply_suggestion / nb_kernel_restart / nb_kernel_restart_and_clear / nb_kernel_interrupt / nb_kernel_list / nb_kernel_select / nb_clear_outputs / nb_set_cwd

nb_add_cell also accepts before (insert above a given cell) in this fork.

Compatibility

DSH 0.1.x-rc and 0.2.x-rc (@deepseek-ai/dsh-tools: ^0.1.0-rc.6 || ^0.2.0-rc.1 || ^0.2.1-rc.1). Developed and tested against DSH 0.2.1-rc.5 on Windows with a conda ipykernel.

License & attribution

MIT. Original work Copyright (c) 2025-present beihzb — @beihaizb/dsh-notebook; modifications Copyright (c) 2026 dsh-jupyterlab contributors. See LICENSE.