A Python sandbox AI feature runs real code in an isolated environment rather than just describing what code would do, which is the difference between an assistant guessing at a calculation and one actually computing it. Ask Mio’s Code runner executes Python or Node to compute something exactly, analyse a table with pandas, draw a matplotlib chart, or convert a file — and it’s one of the clearer examples of why “the AI ran it” beats “the AI wrote code that should work.”
What the Code Runner Actually Does
Per the features page, the Code runner tool “runs Python or Node code in a sandbox to compute something exactly, analyse a table (pandas), draw a chart (matplotlib), or convert a file.” Four distinct jobs live inside that one sentence, and each is worth separating out.
Exact computation
Language models are, fundamentally, next-token predictors — very good at recognizing patterns in text, not built to guarantee arithmetic on large or unusual numbers the way a calculator or a real interpreter does. Ask “what’s 847293 divided by 17, to six decimal places” without a code runner, and a model might get close or might not; run the actual division in Python, and the answer is exact, every time, because it’s real floating-point arithmetic executed by real code rather than a token prediction.
Analysing a table with pandas
Paste a CSV of sales data, or attach a spreadsheet, and ask “what’s the month-over-month growth rate for each product category” — the code runner can load that data into a pandas DataFrame and compute the actual grouped, exact result, rather than eyeballing a table and estimating. This is the same library data analysts use directly, running inside the sandbox on your behalf.
Drawing a chart with matplotlib
Once data is in a DataFrame, generating an actual chart — a bar chart of the category growth, a line chart of a trend over time — is a natural next step, and the code runner can render one as a real image, not a text description of what a chart would show.
Converting a file
Format conversion — turning a CSV into JSON, or restructuring a data file into a different layout — is another concrete job real code handles better than a description of the transformation would.
Why a Sandbox Is Safer Than Running a Stranger’s Script Yourself
If someone sends you a Python script and says “run this, it’ll fix your data,” running it directly on your own machine means that script now has whatever access your user account has — your files, your network, potentially your credentials if the script goes looking for them. A sandbox breaks that link: code runs in an isolated environment with no access to your actual filesystem, accounts or network beyond what’s explicitly provided for the task at hand. If a piece of code turns out to be buggy, wasteful, or even deliberately malicious, the blast radius stops at the sandbox boundary instead of reaching your laptop.
This matters for Ask Mio’s Code runner in both directions. When Mio writes code to solve a problem you asked about, running it in a sandbox rather than describing untested code as a solution means you get a verified result — the code actually ran and produced that output, rather than a plausible-looking snippet that might error the moment you paste it into your own environment. And when you’re pasting in a script from somewhere else and asking Mio to check what it does, running it in the sandbox rather than on your own system is the safer way to find out.
| Approach | Where code runs | Risk if code is wrong or malicious |
|---|---|---|
| Run a script on your own laptop | Your machine, your account permissions | Files, credentials and network access are all exposed |
| Ask an AI to describe what code would do | Nowhere — it’s a prediction, not an execution | No execution risk, but no guarantee the description is correct either |
| Ask Mio’s Code runner | Isolated sandbox | Contained to the sandbox; your own system isn’t touched |
What the Sandbox Is Not
It’s worth being precise here: the Code runner is built for exact computation, table analysis and chart generation as part of answering a question — it is not a general-purpose development environment for building and deploying a full application, and it’s not a substitute for your own tested, reviewed infrastructure when the code in question needs to run in production with real consequences. Treat its output as a verified one-off result or a prototype, not as a component you drop into a live system unreviewed.
Why This Matters More Than It Sounds
It’s tempting to assume a model good enough to write fluent explanations of statistics must also be reliable at doing statistics. The two skills are different in kind. A language model generates text by predicting likely continuations based on patterns in its training; that’s an excellent fit for explaining what a standard deviation means, and a poor fit for guaranteeing the exact standard deviation of a specific 40,000-row dataset it has never seen computed before. A python sandbox AI setup closes that gap by handing the actual arithmetic to an actual interpreter — the model’s job becomes writing correct code and interpreting the result, not performing the calculation itself token by token. This is the same reason human analysts don’t do statistics in their heads for anything that matters: they open a real tool and let it compute, then reason about what the output means. The Code runner gives Mio that same discipline by default for anything numeric.
Sandbox vs a Local Jupyter Notebook
For anyone who already works in Jupyter or a local Python environment, it’s fair to ask what a python sandbox AI feature adds beyond what they could do themselves. The honest answer: nothing computationally — pandas is pandas, matplotlib is matplotlib, the math is the same math. What changes is the path to the result. Instead of opening an editor, writing boilerplate to load a file, writing the actual analysis code, running it, and reading the output, you describe what you want in a sentence and get the computed answer directly, with the code that produced it available if you want to check or reuse it. For someone comfortable writing pandas themselves, the sandbox is a shortcut. For someone who isn’t, it’s the difference between getting an exact answer at all and eyeballing a spreadsheet.
| Situation | Local Jupyter/Python | Ask Mio’s code runner |
|---|---|---|
| One-off exact calculation | Requires writing and running a script yourself | Ask in plain language, get the computed result |
| Environment setup | Requires Python, pandas, matplotlib installed locally | Nothing to install — runs in the sandbox |
| Running someone else’s script safely | Runs with your local permissions unless you sandbox it yourself | Isolated by default |
| Production data pipeline | Appropriate, with proper review and testing | Not the intended use — built for one-off analysis, not deployed infrastructure |
A Worked Example
Say you upload a CSV of a year’s expenses and ask: “which three months had the highest spend, and by how much did November exceed the yearly average?” Without a code runner, an assistant would need to eyeball the numbers in the file and estimate — error-prone once the table has more than a handful of rows. With the code runner, Mio loads the CSV into pandas, computes the actual monthly sums, sorts them, calculates the yearly average, and returns the exact three months and the exact percentage November exceeded it by — a real computed answer, not an estimate, and it can hand back a matplotlib bar chart of all twelve months alongside the number if that’s useful too.
Code Runner vs Code Review, Debug and Refactor
It’s easy to conflate the code runner tool with Ask Mio’s code-related experts and templates, but they solve different problems. The code runner executes code to compute a result as part of answering a question. The “Senior developer” expert reviews, writes and debugs code as an ongoing conversational frame. The “Code review / fix” ready-made job takes a pasted file or ZIP and returns a structured list of bugs and suggested fixes as a deliverable. All three might come up in the same coding session, but only the code runner is actually executing anything.
What a Coding Answer Costs
Ask Mio prices a coding answer at 5–10 points, higher than a normal chat reply’s 1 point or a longer reply’s 3, per the pricing page. A question that routes through the code runner — computing an exact result, analysing a table, or generating a chart — falls into that coding-answer band, reflecting the heavier model and execution work involved compared with a plain conversational reply. The Coding plan’s 6,000 monthly points and 600-per-window allowance is sized with this in mind: a handful of data analysis sessions a day is well within that budget, while very heavy, repeated sandbox use across a team is where the Business plan’s larger allowance and extra seats become the more practical fit.
Frequently Asked Questions
Which languages does the Code runner support?
Python and Node, per the features page — covering both general-purpose scripting and JavaScript-based computation, along with Python’s pandas and matplotlib libraries for data analysis and charting specifically.
Can the code runner access my files automatically?
It works with data you provide directly in the chat — pasted text, an uploaded CSV or spreadsheet — rather than reaching into your filesystem or accounts on its own. That’s a separate concern from connectors, which handle reading your linked accounts, always read-only.
Is running code in the sandbox completely free of any risk?
The sandbox isolates execution from your own machine and accounts, which removes the main risk of running unfamiliar code locally. It’s built for the tasks it’s designed for — computation, table analysis, chart generation, file conversion — rather than as a hardened environment for arbitrary untrusted workloads at scale.
Can the code runner produce an actual chart image, or just describe one?
It can render an actual matplotlib chart as an image, based on real data processed in the same sandbox session — not a text description of what a chart would look like.
Do I need to know Python to benefit from the code runner?
No — you ask your question in plain language (“what’s the average of this column, excluding outliers”), and Mio writes and runs the Python needed behind the scenes. Knowing Python helps you understand or verify the code shown, but isn’t required to get a correct result.
Which plans include the code execution sandbox?
Per the pricing page, the Coding plan (€12/month) explicitly adds Code mode and the code execution sandbox, and the Business plan includes it as well; check current plan details for the exact breakdown.
The Bottom Line
A Python sandbox AI feature turns “the AI wrote code that looks right” into “the AI ran the code and here’s the actual result,” which matters every time a question involves real arithmetic, a real dataset, or a real chart. Ask Mio’s Code runner covers computation, pandas analysis, matplotlib charts and file conversion inside an isolated sandbox that keeps execution off your own machine. If your work involves numbers you need to trust, try asking a data question directly rather than eyeballing a spreadsheet — the Coding plan is where the sandbox lives.