Let’s discuss our work
This release makes notebooks more powerful, more programmable, and more collaborative. From interactive output in Codex and discussions beside the notebook to Jupyter-inspired command shortcuts, more comfortable editing, and improved MCP, CLI, Node, and Python interfaces, nteract 2.7 and 2.8 bring a faster, more flexible way to work with—and build on—your notebooks.
- notebook
- comments
- integrations
Highlights
- Introducing the nteract cli
- Explore the output in Codex
- Make the notebook comfortable to edit
- Inline Commenting
- Embed multiplayer notebooks in your apps
Let’s discuss our work
[I wish I had a decent hook here…]
This release makes notebooks more powerful, more programmable, and more collaborative. From interactive output in Codex and discussions beside the notebook to Jupyter-inspired command shortcuts, more comfortable editing, and improved MCP, CLI, Node, and Python interfaces, nteract 2.7 and 2.8 bring a faster, more flexible way to work with—and build on—your notebooks.
We’ve also got an initial alpha of our cloud notebooks (that you can self-host on CloudFlare or with celld.dev).
Explore the output in Codex
Ask your agent to explore a dataset, then sort and crossfilter the table yourself, right in the conversation. Same with Plotly charts
We’ve had basic MCP Apps support for a while. As the spec has evolved and the harnesses have changed, we’ve had to adapt.
Connect your notebook with the nteract Codex plugin.
Know what your table is showing
Large automatic table outputs show a bounded preview of the first rows. The footer makes clear that sorting, filtering, and summaries apply to those displayed rows. Supported base64 image columns display images instead of a wall of encoded text, including when the first batch contains only empty values.
Changes: #4093, #4153, #4103, #4109
Make the notebook comfortable to edit
Familiar command-mode shortcuts are here. Escape enters command mode; Enter returns to editing. A and B insert a cell above or below, M and Y switch between Markdown and code, D,D deletes a cell, and O toggles output visibility.
Choose code and Markdown fonts with a searchable picker and previews of installed fonts. Turn line numbers on or off. Styled HTML blocks, including nested callouts with backgrounds and borders, render again in Markdown previews.
Changes: #4183, #3960, #3969, #4366
Introducing the nteract cli
Use nteract open analysis.ipynb to open a notebook, nteract . for a fresh notebook in the current directory, or nteract mcp to serve MCP. Notebook operations and workstation commands share one entry point; existing runt and nteract-mcp integrations remain supported.
The CLI-only installer installs the command and helpers without registering or starting a local daemon service. One-shot notebook calls wait for document readiness, and execution calls also wait for the kernel before dispatch.
Inline Commenting
Select any slice of code or markdown (even if rendered!) to leave a note. Same goes for outputs. If you don’t understand a chart, want to change it up, or a dataset needs refinement just leave a note.
As this is a multiplayer feature, it allows for human to human connection as well as agent to human and vice versa.
Hover over a comment highlight to preview the thread. Click to keep it open and reply. Unsent replies survive dismissal, and the Discussions rail lets you browse all threads.
Your conversations stay connected when you save as they’re stored separately and privately on the desktop. When you export a notebook, your discussions stay private and out of the .ipynb file.
When you’re ready to share with people, you’ll need the cloud version.
Changes: #3902, #4182, #4236, #4375
Embed multiplayer notebooks in your apps
We’re going beyond the Desktop app and enabling others to embed nteract in their apps. This starts with our node runtime and electron adapter. An embedding app can connect to the notebook runtime, provide native actions, all without having to rebuild the connection protocol.
Changes: #4134, #4147, #4248, #4287, #4323, #4372, #4285
Environments and platforms
Managed Conda and Pixi environments respect explicit Python 3.12 requests. Dependency approvals distinguish registry packages from requests for another URL, path, or channel.
Windows ARM64 and Linux ARM64 desktop builds join the downloads. Windows artifacts use the static C runtime, removing the dependency on a separately installed Visual C++ Redistributable. On Windows ARM64, managed Conda and Pixi use x64 emulation; native UV environments omit PyArrow from default packages.