Code Executor — Twice The Compute, A Headless Browser, And More Libraries

October 9, 2026

Python runs in Code Executor now get double the CPU and memory, start from a ready-made environment instead of waiting on a rebuild, and come with a headless Chromium browser and a wider set of data, document, and web libraries already installed.

What you can do

• Run heavier analysis. Each Python sandbox now has 4 vCPUs and 8 GB of RAM (up from 2 and 4) and 10 GB of disk (up from 5). Multi-threaded work — numpy and scipy maths, scikit-learn with parallel jobs, xgboost, polars and DuckDB queries — finishes faster, and larger datasets fit in memory.
• Faster numpy and scipy maths. The maths libraries were starting dozens of threads on a few cores and slowing themselves down. They now use one thread per core: in our tests, a large matrix multiply went from 1.1 s to 0.16–0.25 s.
• Start without the wait. The sandbox starts from a pre-built environment, so the first run in a new conversation no longer stalls for a few minutes while the environment rebuilds.
• Render and capture web pages. Playwright with a headless Chromium browser is pre-installed: turn HTML you generated into a PNG or PDF, screenshot a page, or pull content from JavaScript-heavy sites.
• Read and write more formats out of the box: pypdf and pdfplumber for PDF text and tables, reportlab for PDF generation, olefile for legacy .doc and .ppt files, plus markdown and tabulate.
• Crunch data faster: polars, DuckDB (SQL straight over CSV and Parquet), statsmodels (regression and time series), sympy, and networkx.
• Fetch and parse: requests, httpx, and lxml.

Where this shows up

You upload a 2 GB CSV and ask for a breakdown by region. You used to sit through a slow pandas run, or hit the memory ceiling. Now Alfrada can query it with DuckDB or polars on twice the cores.

You built an HTML dashboard and want a PNG of it for a slide. Now Alfrada renders it in headless Chromium and saves the image to your files.

You upload an old .doc from 2003. Alfrada reads it straight away instead of first asking to install a library.

Try it

• "Load in/transactions.csv with DuckDB and give me monthly revenue by region, then chart it."
• "Render out/dashboard.html in a headless browser and save a full-page PNG."
• "Extract every table from in/annual-report.pdf with pdfplumber and put them in one Excel workbook."
• "Fit a seasonal model to in/sales.csv with statsmodels and forecast the next 12 months."

Heads up

• Pulling content from live websites is outbound network access, so it can still pause for your approval in the Safety Center, like any other request from the sandbox.
• The price per run is unchanged.
• A conversation that already has a sandbox from the last 24 hours keeps it until it expires; new conversations get the upgraded one.