Local Python notebooks for a Parquet file that changes: marimo, Deepnote, VS Code and Zed

Choose marimo for a local Python analysis you reopen after the Parquet file changes. Deepnote’s reactive notebooks inside VS Code are a close alternative; VS Code and Zed suit familiar workflows if you rerun before trusting results.

PRACTICAL PICKSPreferredPickOur Preferred PickAn analysis you will reopen and rerun$0

marimo

Free and open source. The notebook is a plain Python file, cells that depend on a change rerun or are marked stale, and an optional file watch reruns them when the data file changes.

The lead over Deepnote is narrow (3.90 versus 3.83). A critical flaw was exploited on exposed servers in April 2026, and only the latest release gets security fixes: keep it updated and on your own machine. CoreWeave owns marimo.

Why marimo is our Pick
Newcomer assessment

No Rising Star selected

Our November 7 assessment found local newcomers, but none established all the availability, notebook and security requirements. Zasper’s release safety and tine’s file persistence remain uncertain.

No Rising Star was selected; choose marimo for this task.

Who is this article for? Python users who keep a small local analysis of a saved Parquet file and need to inspect, chart, save, reopen and rerun it after the file changes upstream.

About 25 min read

What Preferred Pick, Rising Star and Ready to Rock mean

The quick list

Start with marimo for its Python file and data-file watcher. Consider Deepnote for reactive notebooks inside VS Code, VS Code if you already use it, or Zed for ordinary Python cell scripts. Rerun after the Parquet file changes.

The four compared tools have a free local route; project pages checked November 7, 2026, with licence evidence for marimo, VS Code and JupyterLab from October 19. Deepnote’s local tools use Apache-2.0; its cloud plans are separate. marimo is Apache-2.0 licensed. Zed’s personal editor is free without its paid AI features. Visual Studio Code is free under Microsoft’s licence, and its Jupyter extension is MIT-licensed. You install Python and the data packages yourself. Prices for MLJAR Studio’s no-account manual notebook route could not be confirmed.

The comparison at a glance

Compare how each notebook keeps results in step with your code and data, what its file stores, and its 2026 security record. Select up to three tools to compare.

marimo leads narrowly over Deepnote, followed by VS Code and Zed. Choose marimo for the data-file watcher and Python source; all four remain selectable.

Local Python notebooks for a Parquet file that changes
FeaturemarimoDeepnoteVisual Studio CodeZed
Best fitAn analysis you will reopen and rerun as the data changesReactive local notebooks inside VS Code; choose marimo for explicit data-file watchingNotebooks inside the editor you already use for codePython cell scripts in a free editor; choose marimo for dependency tracking and file watching
What the notebook file holdsA .py Python file with code only; last outputs can be kept separately in a __marimo__ folderReadable .deepnote YAML; outputs in separate .snapshot.deepnote files, with code hashesAn .ipynb file as in JupyterLab, or a .py file with # %% cells that holds code onlyOrdinary .py source with # %% cell markers; results appear inline in the editor
On reopeningCells do not run by default; restored outputs are treated as stale until runSnapshots identify the producing code; that does not establish that the Parquet file is unchangedAn .ipynb shows its saved outputs; a # %% file holds no results, which appear in the Interactive window when you run itPython source remains reusable; rerun to produce current results
When the Parquet file changes on diskOptional mo.watch.file reruns dependent cells; otherwise run againAn equivalent to mo.watch.file is not established; rerun after replacing the dataNo built-in file watch; run againNo data-file watcher established; rerun after replacing the file
2026 security recordCritical sign-in bypass (fixed in 0.23.0) exploited within hours and listed by CISA; reflected XSS fixed in 0.23.9No published advisories on the project security page; local extension coverage is not established22 VS Code advisories on September 8 and October 13, two critical, one through a malicious .ipynb; all fixed in 1.141.1Advisories include untrusted project, LSP and extension archive code execution; current patched versions not established
Main limitationA different cell model to learn, a 2026 exploited flaw, and a single corporate ownerExtra extension and toolkit; code hashes do not prove the data is current; YAML and publishing issue reportsNo protection against stale results beyond your own rerun; a heavy recent advisory recordManual reruns, separate kernel setup, uncertain advisory patch mapping; telemetry on by its privacy page
Reader reviewsNo reviews yet. Write a reviewNo Practical Picks reader reviews yetNo reviews yet. Write a reviewNo Practical Picks reader reviews yet

marimo leads narrowly over Deepnote (3.90 versus 3.83), followed by VS Code and Zed. Choose marimo for its Python source and explicit data-file watcher. No product was installed or run for this review; the one row from our October 1, 2026 check is labelled as historical. Only marimo has an explicit data-file watcher established by the documentation we read, and only if you wrap its path in mo.watch.file. Otherwise run the notebook again and check the expected rows and totals before you rely on or share a result.

All features and sources 15 criteria

Open the groups that matter to you. They compare the same selected tools. “Planned” means future work, not an available feature. Sources include the published conditions.

Keeping results current

Show 5 criteria
Keeping results current
FeaturemarimoDeepnoteVisual Studio CodeZed
On reopeningCells do not run by default; restored outputs are treated as stale until runSnapshots identify the producing code; that does not establish that the Parquet file is unchangedAn .ipynb shows its saved outputs; a # %% file holds no results, which appear in the Interactive window when you run itPython source remains reusable; rerun to produce current results
When a cell or input changesCells that use the changed variables rerun automatically, or are marked stale in lazy mode; in-place changes to objects are not trackedMaker documents automatic reruns of dependent blocks when inputs changeRerun cells yourself: Run All, Run Above or Run BelowRun cells or selections yourself; no dependency tracking established
When the Parquet file changes on diskOptional mo.watch.file reruns dependent cells; otherwise run againAn equivalent to mo.watch.file is not established; rerun after replacing the dataNo built-in file watch; run againNo data-file watcher established; rerun after replacing the file
Rerun without the editorpython notebook.py, or marimo export to HTMLLocal toolkit and CLI; no editor required for the documented execution toolsA # %% file runs with python; an .ipynb needs nbconvert or similarThe ordinary Python source can run as a script with the same dependencies
Earlier hands-on check, October 1, 2026 (before our research-only method)0.25.0: updated totals after reopening and Run All; a slider change reran dependent cells; replaced file not noticed by itselfNot part of that checkNot part of that checkNot part of that check
Sources for keeping results current

Zed Python REPL and cell scripts, pricing, privacy, terms and security advisories, checked November 7, 2026. Deepnote local execution and format, checked November 7, 2026. marimo reactivity, file watching, scripts and 0.11.7 notes on stored outputs; JupyterLab commands, the notebook file format and nbconvert execution; VS Code notebooks and Python Interactive window. Checked October 19, 2026. These are documented behaviours, not test results, except the earlier-check row: a dated historical observation on one synthetic eight-row file, made before we adopted our research-only method; see the earlier check record.

Files, setup and data

Show 5 criteria
Files, setup and data
FeaturemarimoDeepnoteVisual Studio CodeZed
Best fitAn analysis you will reopen and rerun as the data changesReactive local notebooks inside VS Code; choose marimo for explicit data-file watchingNotebooks inside the editor you already use for codePython cell scripts in a free editor; choose marimo for dependency tracking and file watching
Cost and licence$0; Apache-2.0; owned by CoreWeave since October 2025$0 local tools; Apache-2.0; hosted collaboration separate$0; Microsoft licence for the editor; Jupyter extension MIT$0 personal editor, anonymous use; AI plans are optional
What the notebook file holdsA .py Python file with code only; last outputs can be kept separately in a __marimo__ folderReadable .deepnote YAML; outputs in separate .snapshot.deepnote files, with code hashesAn .ipynb file as in JupyterLab, or a .py file with # %% cells that holds code onlyOrdinary .py source with # %% cell markers; results appear inline in the editor
Looking at the dataDataframes shown as searchable, sortable tables; SQL cellsPython and SQL blocks, input and visualization block types; ordinary Python data librariesVariables view and Data Viewer for dataframesPython kernel outputs displayed inline; use your Python dataframe and plotting libraries
Local server and accessBrowser editor on 127.0.0.1 with an access token by defaultLocal editor extension and Python toolkit; cloud upload and sharing are separate featuresNo browser notebook server; the extension starts the kernel itselfEditor REPL talks to Jupyter kernels; Python and ipykernel installed separately
Sources for files, setup and data

Zed Python REPL and cell scripts, pricing, privacy, terms and security advisories, checked November 7, 2026. Deepnote local tools and format and company history, checked November 7, 2026. Package records for marimo and JupyterLab; marimo dataframes and authentication; Jupyter Server security; VS Code licence, Jupyter extension licence and its kernel design notes. Checked October 19, 2026.

Security, maintenance and ownership

Show 5 criteria
Security, maintenance and ownership
FeaturemarimoDeepnoteVisual Studio CodeZed
2026 security recordCritical sign-in bypass (fixed in 0.23.0) exploited within hours and listed by CISA; reflected XSS fixed in 0.23.9No published advisories on the project security page; local extension coverage is not established22 VS Code advisories on September 8 and October 13, two critical, one through a malicious .ipynb; all fixed in 1.141.1Advisories include untrusted project, LSP and extension archive code execution; current patched versions not established
Security fixes forThe latest release only, by its own policyActive package releases; no local long-term security support promise establishedThe current release, usually updated weeklyNo SECURITY.md detected on the project security page; support window unknown
Latest release0.25.2, October 19, 2026; a release every one to three weeks; still before 1.0Reactivity package 2.0.0, November 5, 2026; other components have separate versionsVS Code 1.142, October 14, 2026, most weeks; Jupyter extension build of July 15, 2026Current stable version not established from the inspected pages; do not infer one from an advisory date
Who stands behind itMarimo Inc., a CoreWeave subsidiary; team says it stays free and open sourceDeepnote Inc.; notebook company founded in 2019; core opened in November 2025MicrosoftZed Industries Inc., named in its terms; independent entity confirmation not established here
Main limitationA different cell model to learn, a 2026 exploited flaw, and a single corporate ownerExtra extension and toolkit; code hashes do not prove the data is current; YAML and publishing issue reportsNo protection against stale results beyond your own rerun; a heavy recent advisory recordManual reruns, separate kernel setup, uncertain advisory patch mapping; telemetry on by its privacy page
Sources for security, maintenance and ownership

Zed Python REPL and cell scripts, pricing, privacy, terms and security advisories, checked November 7, 2026. Deepnote releases, security reporting and issue list, checked November 7, 2026. Security advisories for marimo, JupyterLab, Jupyter Server and VS Code; the CISA Known Exploited Vulnerabilities catalog; security policies of marimo; release histories of marimo, JupyterLab and VS Code; marimo’s CoreWeave announcement and Project Jupyter’s move to LF Charities. Checked October 19, 2026.

What matters when choosing

marimo is our Preferred Pick for a Python user who keeps a small local analysis of a Parquet file that changes upstream, and needs to know that a reopened result reflects the current file. It is free and open source under the Apache 2.0 licence. A marimo notebook is a plain Python file that holds your code, not your results, and marimo tracks which cells use which variables: when a cell or an input changes, every cell that depends on it runs again, or is marked stale if you choose the lazy setting. It does not watch your data file unless you ask it to. Wrap the file’s path in mo.watch.file and the cells that read it rerun when the file changes; otherwise, run the notebook again after the file changes. The same file also runs as an ordinary script with python notebook.py, so you can repeat the whole analysis outside the editor. Reactivity, file watching, running as a script, package record

The lead over Deepnote’s local edition is narrow: 3.90 against 3.83 out of five. Choose marimo for its ordinary Python file and explicit data-file watching; consider Deepnote if you want reactive notebooks inside VS Code.

Two cautions belong beside that recommendation. The first is security. In April 2026 a critical flaw let anyone who could reach a marimo server open a shell on it without the access token. Researchers at Sysdig saw it exploited less than ten hours after disclosure, and the US Cybersecurity and Infrastructure Security Agency (CISA) lists it as exploited. It was fixed in version 0.23.0, and marimo issues security fixes only for its latest release, so update often, keep the server on your own machine (127.0.0.1, its default) and leave the access token on. The second is ownership. CoreWeave, a cloud-computing company, bought Marimo Inc. in October 2025. The team says marimo will stay free and open source with the same roadmap; that is a promise, not a guarantee, although the Apache licence lets anyone continue the code. Expect a learning curve as well: each variable can be defined in only one cell, and marimo does not notice when an object is changed in place. Advisory, InfoWorld on Sysdig’s findings, April 13, 2026, CISA catalog, security policy, CoreWeave announcement

Consider JupyterLab if you share notebooks with people who use Jupyter, or rely on its extensions. It is the established, community-governed notebook workspace, and its .ipynb file is what most tools and colleagues expect. That file stores outputs alongside your code, so a reopened notebook shows the totals from its last run, with nothing to tell you whether the data has changed since. Use Restart Kernel and Run All before you trust or share a result. If you already work in Visual Studio Code, consider its Jupyter extension. It runs the same .ipynb notebooks, and its Python Interactive window runs # %% cells from an ordinary .py file whose results are not saved into the file. It has no dependency tracking either, and Microsoft published 22 VS Code security advisories in September and October 2026, one of them reachable by opening a malicious notebook. JupyterLab commands, notebook file format, VS Code Interactive window, VS Code notebook advisory

These are researched judgments from documentation, release records, security advisories and public reports: evidence for marimo, VS Code and JupyterLab was checked October 19, 2026, with Deepnote and additional alternatives examined November 7, 2026. We did not install or run any notebook for this review. In an earlier hands-on check on October 1, 2026, before we adopted our research-only method, marimo 0.25.0 and JupyterLab 4.6.4 both reached the corrected totals from a small synthetic Parquet file once each notebook was run again after the file changed; the reopened JupyterLab notebook showed the old total until then. That check is consistent with the documentation; it did not decide the Pick. Whichever notebook you use, your Parquet file and your code stay ordinary files you can open elsewhere, and none of these tools is a substitute for checking the expected row count and totals after the data changes. Earlier check record, October 1, 2026

Disclosure: Practical Picks articles are researched and written with AI, currently Anthropic’s Claude. marimo offers Anthropic as an AI provider and documents use with Claude Code; Jupyter AI, Project Jupyter’s AI extension for JupyterLab, lists Claude among the agents it connects to; Visual Studio Code offers Anthropic’s Claude as a built-in model provider; and Zed advertises Anthropic models. We have no relationship with marimo, CoreWeave, Project Jupyter, Microsoft, Zed or Anthropic over this article, and these Claude links carried no weight in our judgment. marimo AI, Jupyter AI, VS Code third-party agents, Zed AI and privacy, checked November 7, 2026

marimo

Preferred Pick · rerunnable analysis

$0Free, open source (Apache-2.0) · owned by CoreWeave

Choose marimo for a local analysis you will reopen and rerun as the Parquet file changes, and keep it updated.

Reasons to choose it

  • Notebook is a plain Python file that holds code, not results
  • Cells that depend on a change rerun or are marked stale
  • Optional file watch reruns cells when the data file changes

What to watch

  • A critical 2026 flaw was exploited on exposed servers; fixes only for the latest release
  • A different cell model to learn
  • Owned by one company since October 2025

marimo is a Python notebook that runs in your browser from a local server, started with marimo edit; an official VS Code extension and an early-preview PyCharm plugin open the same notebooks in those editors. We verified that version 0.25.2, released October 19, 2026, requires Python 3.10 or later and is Apache-2.0 licensed. It does not include a Parquet reader: install pandas with PyArrow, Polars or DuckDB yourself, or declare them inside the notebook so that uv or Pixi can build a separate environment for it. Dataframes appear as tables you can page through, search, sort and filter, and SQL cells are built in. Package record, dataframes, scripts and declared dependencies, own editor and extensions, PyCharm plugin, July 21, 2026

Its useful difference is how it keeps results consistent with code. marimo reads each cell to see which variables it defines and which it uses, and builds a graph from that. Run a cell and every cell that uses its variables runs too; in lazy mode they are marked stale instead. Two limits matter. marimo does not track changes made to an object in place, such as setting an attribute, and a variable may be defined in only one cell. By default, opening a notebook runs nothing: marimo stores the last outputs in a __marimo__/session folder and brings them back when you reopen the notebook, still treated as stale until you run the cells. Session snapshots made with marimo export session are rewritten only when marimo considers them stale, for example after the code or its declared dependencies change; the documentation does not say a data change counts, so pass --force-overwrite after the Parquet file changes. Reactivity, 0.11.7 notes on stored outputs, session snapshots

Read the full marimo analysis and sources

For a file that changes underneath you, mo.watch.file wraps a path so that cells reading it rerun when the file changes; mo.watch.directory reacts only when files are added or removed. In an earlier hands-on check on October 1, 2026, before we adopted our research-only method, marimo 0.25.0 did not notice our replaced eight-row Parquet file by itself, because the check did not use mo.watch.file; after reopening and Run All it showed the updated totals, and moving a slider in the notebook reran the dependent total without Run All. Watch API, earlier check record, October 1, 2026

Security is the main reason for care. The critical flaw published on April 8, 2026 (CVE-2026-39987) was an unauthenticated terminal connection that gave a shell to anyone who could reach the server; it affected versions before 0.23.0. Sysdig reported an attacker reading cloud keys from one of its deliberately exposed test servers within ten hours of disclosure, and advised treating any publicly reachable instance as compromised. CISA added the flaw to its catalog of exploited vulnerabilities on April 23, 2026. A June advisory fixed a reflected cross-site scripting flaw in 0.23.9, and an October 2025 advisory closed an unauthenticated proxy route in 0.16.4. marimo’s policy is to fix only the latest stable release. The editor listens on 127.0.0.1 with an access token by default; do not turn the token off or expose the port to a network. Critical advisory, cross-site scripting advisory, proxy advisory, InfoWorld, security policy, authentication

marimo is young and owned by one company. Marimo Inc. raised a $5 million seed round in November 2024, with AIX Ventures leading, and CoreWeave bought it on October 30, 2025, when the team numbered eight. The announcement says marimo “is, and always will be, free, open-source, and permissively licensed” and that the open-source roadmap is unchanged. Releases are frequent, every one to three weeks, and the project is still before version 1.0: 0.24.0 kept its file-browser widget inside its starting folder by default, and 0.25.2 stopped switching plot themes to match a dark notebook. An open report from 2025 describes a notebook under /tmp that could not be saved in version 0.14.13. Choose JupyterLab instead if colleagues expect .ipynb files, although marimo can export to that format and convert from it. Seed announcement, CoreWeave announcement, releases, save report, coming from Jupyter

Used marimo? Share your experience to help other readers choose.

Deepnote

Also consider · reactive notebooks in VS Code

$0Local tools: Apache-2.0; cloud service separate

Consider Deepnote for reactive local notebooks in VS Code; choose marimo for its Python file and explicit data-file watcher.

Reasons to choose it

  • Dependent blocks rerun automatically
  • Readable YAML with outputs kept in separate snapshots
  • Converts to and from Jupyter notebooks

What to watch

  • Data-file watching is not established by the documentation
  • Editor extension and Python toolkit to maintain
  • Public reports include YAML parsing and publishing problems

Consider Deepnote’s local edition if you want dependent blocks to rerun inside VS Code, Cursor or Windsurf. Its maker documents local Python, R and SQL execution, automatic reruns of dependent blocks, and conversion to and from Jupyter notebooks. The local tools and format are Apache-2.0 licensed; the hosted collaboration service is a separate product. We checked these sources November 7, 2026. Local project, company and editions

A .deepnote project is readable YAML. Separate .snapshot.deepnote files hold outputs, with a code hash identifying the code that produced them. That makes source changes easier to review than an .ipynb containing all its outputs, but a code hash does not prove that the Parquet file is unchanged. The documentation we read does not establish an equivalent to marimo’s explicit data-file watcher: rerun after replacing your data and check the totals. The editor extension and Python toolkit are additional parts to install and maintain. Format and tools, checked November 7, 2026

Read the full Deepnote analysis and sources

The public issue list includes reports about valid YAML being rejected and publishing commands overwriting app source or moving directories into another project. Those publishing paths are outside this local analysis, but they are reasons to keep backups and check what a command will write. An issue title is a report, not a verified defect in every release. The repository’s security page lists no published advisories; that does not establish that it has no vulnerabilities. Cloud security certifications and encryption claims do not establish the safety of the local extension. Issue list, security reporting, cloud security, all checked November 7, 2026.

Deepnote began as a cloud notebook company in 2019 and opened its core in November 2025. It is an established notebook maker, so its local edition does not qualify for our Rising Star award. Its local execution and readable format still make it a credible alternative here. Company history, checked November 7, 2026

Used Deepnote? Share your experience to help other readers choose.

Visual Studio Code

Also consider · already in VS Code

$0Free · Microsoft licence; Jupyter extension MIT

Consider VS Code if you already write code there; use # %% Python files when you want code-only files.

Reasons to choose it

  • Runs .ipynb notebooks and # %% cells in ordinary .py files
  • Variables view and Data Viewer for dataframes
  • No browser notebook server

What to watch

  • No dependency tracking or file watch
  • 22 security advisories in September and October 2026
  • Telemetry on by default; Copilot built in

Visual Studio Code is Microsoft’s free code editor. Its Jupyter extension, which the Marketplace counts at over 110 million installs, opens and runs .ipynb notebooks with Run All, restart, a Variables view and a Data Viewer for dataframes; you also need a Python environment with the ipykernel package. The extension starts the kernel itself rather than through a browser notebook server. The editor ships under Microsoft’s own licence; the Jupyter extension is MIT-licensed. Its newest Marketplace build is dated July 15, 2026, while its GitHub release tags stop at October 2025, so watch its maintenance. Jupyter notebooks in VS Code, Marketplace listing, kernel design notes, VS Code licence, extension licence

VS Code offers two ways to do this job. An .ipynb notebook behaves as in JupyterLab: saved outputs reappear when you reopen it, and nothing tracks which cells depend on which. The alternative is an ordinary .py file divided into cells with # %% comments and run in the Python Interactive window. That file holds only code and runs unchanged as a script; results appear in the Interactive window rather than in the file, and you can export them to a notebook with outputs when you need a record. In a folder you have not trusted, VS Code will not run notebook cells. Neither path watches your data file. Python Interactive window, Jupyter notebooks in VS Code

Read the full Visual Studio Code analysis and sources

Keep it updated. Microsoft published 11 VS Code security advisories on September 8 and 11 more on October 13, 2026, two of them critical and several describing ways around Workspace Trust, the protection for folders you have not trusted. One, fixed in 1.136.2, let a malicious .ipynb notebook run JavaScript that could read and send out local files even in an untrusted folder. All 22 are fixed in 1.141.1. Usage telemetry is on until you set telemetry.telemetryLevel to off, and GitHub Copilot’s chat and suggestions are built in; chat.disableAIFeatures hides them. Notebook advisory, Workspace Trust advisory, VS Code advisories, telemetry

Choose VS Code when you want notebooks next to the rest of your code and already keep the editor updated. If you want marimo’s dependency tracking without leaving VS Code, marimo publishes its own VS Code extension.

Used Visual Studio Code? Share your experience to help other readers choose.

Zed

Also consider · Python cell scripts

$0Free personal editor; paid AI optional

Consider Zed for ordinary Python cell scripts; choose marimo for dependency tracking and a data-file watcher.

Reasons to choose it

  • Runs # %% Python cells through Jupyter kernels
  • Source remains an ordinary Python file
  • Free without an account

What to watch

  • No established dependency tracking or data-file watcher
  • Current patched versions for listed advisories are uncertain
  • Separate Python/kernel setup; telemetry on by its privacy page

Zed runs Python through Jupyter kernels inside ordinary source files. Divide a .py file with # %% markers and run a cell, selection or line; results appear inline and can be cleared. Choose it if you want an editor centred on reusable Python scripts rather than saved notebook outputs. The editor is free without an account; paid AI features are optional for this task. Python REPL and cell mode, pricing, anonymous use, checked November 7, 2026.

You still need a Python environment and ipykernel. On macOS, the documentation says the system Python will not work; use a virtual environment or pyenv, and align the editor’s Python toolchain with the selected kernel. The sources do not establish dependency tracking or an automatic watcher for a changed Parquet file. Run the analysis again and check its final totals after replacing the data. Environment and kernel setup, checked November 7, 2026.

Read the full Zed analysis and sources

Security is a reason for care. Zed’s advisory list includes code execution through untrusted repositories, language-server configuration and extension archives, as well as problems with agent permissions. The list we read does not establish which current releases contain every fix, and GitHub detected no security policy. Our security judgment is therefore less certain than its documented cell workflow. Keep the editor updated and treat unfamiliar projects and extensions as executable code. The privacy page says both that collection requires consent and that telemetry is on; assume it is on until you disable it. Security advisories, privacy and telemetry, checked November 7, 2026.

Zed scores 3.25 out of five for this job, behind VS Code’s 3.30. Choose marimo for dependency tracking and an explicit data-file watcher, or VS Code if you need its documented .ipynb workflow and dataframe viewer. Zed’s plain scripts remain useful without giving it a newcomer award.

Used Zed? Share your experience to help other readers choose.

Which product are we missing?

Know a product that belongs in this article? Tell us what it does differently and why it deserves a closer look. Newcomers welcome.

Readers and product teams can suggest candidates. Suggestions inform our research; they do not guarantee inclusion or a recommendation.

Suggest a product

Opens an email draft with this article attached as a link. Nothing is sent until you send it.

No email app? Send the product name, website, and this article’s title to admin@practicalpicks.guide.

What people are saying

There are no Practical Picks reader reviews for these tools yet. The clearest independent evidence about marimo is the April 2026 exploitation reported by Sysdig and covered by InfoWorld: on servers Sysdig had deliberately left exposed and unpatched, an attacker found and read a file of cloud credentials in under three minutes, and Sysdig advised rotating any keys stored on servers that had been publicly reachable. InfoWorld reported that CoreWeave did not immediately respond to a request for comment. InfoWorld, April 13, 2026

On saved notebooks in general, a 2019 study of 1.4 million Jupyter notebooks on GitHub found that only 24.11% of 863,878 attempted runs finished without errors and only 4.03% reproduced their saved results; it named missing dependencies, hidden state, out-of-order execution and inaccessible data among the causes. It measured public notebooks, not a careful analyst’s own work, but it shows why a saved .ipynb result should be rerun before it is trusted. Pimentel and others, MSR 2019

Issue trackers show bounded leads, not frequencies. A marimo report that a notebook under /tmp could not be saved in version 0.14.13 remains open; a report of a cached error blocking a notebook is now closed. A JupyterLab saving report from version 3.4.3 with real-time collaboration is closed. Our October 19 search for vulnerabilities, outages, lawsuits and ownership changes involving marimo, VS Code and JupyterLab found nothing else that changed our conclusions; a search that finds nothing is not proof that nothing exists. marimo save report, marimo cached-error report, JupyterLab save report

How we researched

This is public-source research. For this review we did not install or run any notebook, create accounts or contact any project. No project supplied access, payment or input, and we use no affiliate links. We first compared marimo and JupyterLab on October 1, 2026, with a small hands-on check on synthetic data made before we adopted our research-only method; that record stays available as dated historical evidence. On October 19, 2026 we rechecked every material claim against current releases, documentation, licences, security advisories, the CISA catalog, ownership records and public reports, and added Visual Studio Code.

We wrote this for a Python user who keeps a small local analysis of a saved Parquet file: inspect it, total and chart it, save the notebook, and reopen and rerun it after the file changes upstream. Before scoring, we fixed five criteria and their weights: protection against stale results when code or data change (30%), a reviewable, rerunnable notebook file (20%), local setup and working with the data (15%), security and maintenance record (20%), and familiarity, ecosystem and continuity (15%). marimo leads, narrowly ahead of Deepnote’s local edition; VS Code and Zed follow. AI features carried no weight.

On November 7, 2026 we examined public project, release and issue records for additional local options. Deepnote’s local tools merit comparison even though its maker is an established notebook vendor. Zed’s documented Python cell scripts place it fourth, at 3.25. JupyterLab and the simpler Notebook 7 both score 3.20: their .ipynb format is useful for sharing, but saved outputs need a fresh run after the data changes. JupyterLab remains worth considering for its file browser, terminal, extensions and community governance; our October 19 assessment found its published 2026 advisories fixed in 4.6.4. Positron scores 3.13, with a useful Parquet explorer but saved notebook outputs. RStudio scores 3.00, PyCharm 2.95, nteract 2.88 and Zeppelin 2.58. These are judgments for this small local task, not general quality ratings. JupyterLab, Notebook, Positron, RStudio, PyCharm editions, nteract, Zeppelin Python; additional alternatives checked November 7, 2026.

No Rising Star was selected in our November 7, 2026 assessment. Zasper’s latest listed release predates source changes that fix reachable vulnerabilities; we could not establish the severity and applicability of those flaws well enough to recommend it. tine has local Windows and macOS releases, but its notebook persistence and security evidence did not meet our requirements. Data Studio, Clusy, Jupyt and Stravu lacked identifiable current product documentation sufficient for this task. Deepnote and Positron come from established notebook or data-science IDE vendors, nteract is an established product, MLJAR Studio first appeared in 2021, and marimo’s extensions come from the Pick’s maker. Zed’s qualifying first-release date was not established. None earns the newcomer award merely through a relaunch, a small audience or AI features. Zasper releases, security changes, tine, MLJAR Studio, checked November 7, 2026; historical launch dates are from our October 19 research.

Spyder, JupyterLab Desktop, Zasper, MLJAR Studio, Data Studio and tine remain unscored because current dependency safety, file behaviour or core-task documentation could not be established sufficiently. Spyder’s separate notebook plugin also warns against pip installation and does not support its standalone installers. Prices for MLJAR Studio’s no-account manual notebook route could not be confirmed. PyCharm’s free core now includes basic notebook execution and output, though we did not reconfirm account-free activation. Google Colab, Hex, Deepnote Cloud and molab are hosted services; DuckDB UI uses SQL rather than Python. Spyder plugin, JupyterLab Desktop security, MLJAR pricing, molab, checked November 7, 2026.

ToolPublic sourcesWhat remains uncertain
marimoPackage record, documentation on reactivity, file watching, scripts, session snapshots, authentication and dataframes, release notes, security policy and advisories, CISA catalog, InfoWorld report, seed and CoreWeave announcements, issue tracker.How the session-output folder behaves in every version, whether the remaining terminal and proxy routes are hardened beyond the published fixes, CoreWeave’s long-term priorities.
ZedPython REPL and cell-mode documentation, pricing, privacy and security-advisory pages, checked November 7, 2026.Current patched versions for listed advisories, runtime support, and telemetry defaults given conflicting statements.
DeepnoteLocal source and format description, releases, issue list, security reporting, company history, pricing and legal pages, checked November 7, 2026.Data-file watching, local extension security coverage, and the scope of reported YAML and publishing failures.
Visual Studio CodeJupyter and Interactive window documentation, Marketplace listing, extension licence and kernel design notes, VS Code licence, telemetry documentation and advisories.Whether the Jupyter extension’s newest Marketplace build is a stable release, how actively the extension is maintained, telemetry from extensions.
Newcomer assessmentNovember 7 project and release records, plus dated October 19 launch records.No candidate established every availability, task and safety requirement for a newcomer award.

None of this establishes speed, behaviour on large files, crash recovery or security on your machine. Next substantive review is due December 18, 2026, or sooner after a security advisory or major release for any of the four. Found a changed release or behaviour? Contact the editors.

Sources and evidence

Evidence for marimo, VS Code and JupyterLab was checked October 19, 2026; the additional sources dated November 7, 2026 cover alternatives and current price information. Project material establishes documented behaviour and published releases, not behaviour in your environment. Security advisories describe affected versions and conditions as their publishers state them. The earlier check is a dated historical observation, not a current test.

This is a researched comparison, not a product review. Method: evidence standard, editorial picks.

No screenshots are shown. This comparison is based on public documentation, release records and security advisories; an earlier hands-on check on synthetic data, made on October 1, 2026 before we adopted our research-only method, is kept as a dated record.