"Data science" used to mean opening a Jupyter notebook, remembering which cell you ran last, and writing pandas code from memory. Two things have changed that: notebooks that no longer let you shoot yourself in the foot with hidden state, and AI tools that let you just type a question in plain English and get back a chart or a number instead of writing the query yourself. This guide covers both kinds — next-generation notebooks for people who still want to write code, and AI analysts for people who'd rather not. Seven real tools, what each actually does, and where it falls short.
marimo — the notebook that fixes Jupyter's biggest annoyance
marimo is a reactive Python notebook stored as a plain .py file instead of a JSON blob. Run a cell, and every cell that depends on it re-runs automatically — there's no more "wait, did I re-run cell 3 after I changed cell 1?" ghost-state bug that's cost every Jupyter user an afternoon at some point. Because it's just a .py file, it's Git-diffable and can be deployed straight as a web app or compiled to WASM. It's free and open source (Apache 2.0), full stop.
Good for: anyone who's been burned by Jupyter's out-of-order execution and wants notebooks that behave like actual programs. Limits: existing Jupyter widgets don't carry over, and migrating old notebooks takes real refactoring work.
Julius AI — talk to your spreadsheet like a person
Julius AI lets you upload a CSV, Excel file, or even a PDF and ask questions about it in plain English — "what's driving the drop in March?" — and get back a chart, a trend line, or a forecast. Under the hood it's actually writing and running Python/R code, not faking it, so the analysis is real and reproducible. The free tier caps out at 15 messages a month, which is enough to try it and not much else; Plus runs $35/month for 250 messages, Pro is $45/month for unlimited messages plus database connectors (MySQL, PostgreSQL, BigQuery).
Good for: non-technical people who need real statistical answers without learning pandas. Limits: the free tier is too thin for ongoing use, and it's less built for team collaboration than a dedicated BI tool.
Powerdrill — an analyst that remembers your last conversation
Powerdrill does the same plain-English data Q&A as Julius, with one difference: it keeps context across sessions, so it gets sharper about your specific data the more you use it, and ships prebuilt "explorers" for economic, financial, and healthcare datasets. The free tier gives 1,000 daily credits, which is generous enough to actually test it; paid tiers range from roughly $13/month (Pro) to $166/month (Premium).
Good for: people who want a no-code AI analyst that improves with repeated use, not a one-off question tool. Limits: the four-tier credit pricing takes real effort to map to your actual usage, and daily credit caps can feel tight if you're in it every day.
Zerve — an AI notebook built for enterprise data teams
Zerve is a data-science notebook where the AI agent already has context on what's in your data warehouse and retains what your team learned in past sessions — no starting from a blank cell every time. Notebooks deploy directly to an API, app, or dashboard in one click, and there are on-premise and air-gapped options for sensitive data. It counts Airbus, NASA, IBM, and Tesco as customers. Free tier gives 150 credits/month across 4 editors; Pro is $18.75/user/month, Team $37.50/user/month, both billed annually, with credits charged at LLM API cost plus 20%.
Good for: larger teams that want an enterprise-grade AI notebook with on-prem deployment options. Limits: credit-based pricing (API cost + 20%) is less predictable than a flat seat price, and it's overkill for solo, occasional use.
SciStudio — a free canvas for scientists juggling five tools
SciStudio tries to put notebooks, plotting, and data pipelines on one collaborative canvas instead of forcing researchers to bounce between five separate apps. It's completely free and open source, hosted on GitHub, with an active Slack/Discord community. Being young shows: the site still has some placeholder content and documentation is minimal.
Good for: researchers who want a free, open-source alternative to juggling JupyterLab plus a separate plotting tool. Limits: very early-stage — thin docs, an uncertain funding model, and what looks like a small team or solo maintainer behind it.
Autoplot — publication-ready charts without leaving your Mac
Autoplot is a native Mac app that takes a spreadsheet of numbers and turns it into a publication-ready chart, built specifically for researchers and students tired of juggling a spreadsheet, a plotting script, and a separate export step. It bundles scientific analysis tools — power-law fits, CCDF, FSS, correlation matrices — plus an AI assistant that writes and runs Python from a plain-language description. It's local-first and works offline. Free tier gives 10 AI credits/month; Plus is $7.99/month, Pro $17.99/month for 1,000 credits.
Good for: Mac-using researchers and students who want clean figures for a paper without fighting matplotlib. Limits: Mac-only, no cloud collaboration, and less flexible than a hand-written script for truly custom visualizations.
MinusX — an AI analyst that sits on top of your existing BI tool
MinusX answers plain-English questions about your business data — a chart, a number, an explanation — instead of you waiting on the data team to write SQL. It ranked #1 on UC Berkeley's independent DataAgentBench benchmark, and there's a genuinely free, self-hosted open-source version if you're willing to manage your own LLM keys. Hosted plans start at $40/user/month (Founders, 500 credits) up to $600/month for a 20-user Team plan.
Good for: teams that already have a data warehouse and want an AI layer on top of it, with a real open-source escape hatch. Limits: needs an existing warehouse to connect to, and the self-hosted route requires real technical setup and LLM key management.
Side by side
| Tool | Best for | Price | Platform |
|---|---|---|---|
| marimo | Reactive, bug-resistant Python notebooks | Free, open source | Web / local, deploys to WASM |
| Julius AI | Chat with your spreadsheet | Free (15 msg/mo); $35–$45/mo | Web app |
| Powerdrill | AI analyst that improves over time | Free (1,000 credits/day); $13–$166/mo | Web app |
| Zerve | Enterprise AI notebook, on-prem option | Free (150 credits/mo); $18.75–$37.50/user/mo | Web, on-prem/air-gapped |
| SciStudio | Free unified research canvas | Free, open source | Web |
| Autoplot | Publication-ready charts, offline | Free (10 credits/mo); $7.99–$17.99/mo | Mac only |
| MinusX | AI layer over your existing warehouse | Free (self-hosted); $40–$600/mo hosted | Web, self-hostable |
If you still want to write Python but you're tired of Jupyter's hidden-state bugs, marimo is a straightforward, free swap. If you'd rather skip code entirely, Julius AI and Powerdrill both do plain-English analysis — Julius leans toward one-off statistical questions, Powerdrill toward an assistant that gets better the more you use it. Teams with an existing data warehouse and a need for governance should look at Zerve or MinusX, the latter with a real open-source option. And if you're a researcher who just wants clean output — a canvas or a chart — without enterprise overhead, SciStudio and Autoplot cover that at little to no cost.