A partial snapshot of the isolated Linux execution environment used by ChatGPT (OpenAI) when processing requests with the code interpreter / computer use feature. Extracted from a live container session and published for research and comparison purposes.
| Parameter | Value |
|---|---|
| OS | Ubuntu 22.04.3 LTS (Jammy Jellyfish) — Debian 12 base packages |
| Kernel | Linux 5.15.0 (Azure infrastructure) |
| User | oai (home: /home/oai) |
| Writable storage | /mnt/data |
| Container runtime | OpenAI CAAS (Container As A Service) infrastructure |
Shell config (.bashrc, .profile):
Standard Ubuntu bash setup with NVM (Node Version Manager) sourced at the end — indicating Node.js is available via nvm.
redirect.html — A small HTML redirect page present in the home directory.
.wgetrc — wget configuration file, likely used for downloading files within the container.
.chromium/ — Chromium browser profile (Default/Preferences, Local State), confirming a headless Chromium browser is present for web browsing tasks.
.config/openbox/rc.xml — Openbox window manager config, part of the desktop environment stack (Xvfb + Openbox + XFCE4).
.ipython/ — IPython profile with startup config and history.sqlite — the Jupyter/IPython kernel history database.
These are markdown and Python files that instruct the model how to perform specific tasks reliably. Similar in concept to Claude's skills system.
| Skill | Files | Description |
|---|---|---|
docs |
skill.md, render_docx.py |
Create and review Word documents using python-docx + LibreOffice render loop |
pdfs |
skill.md |
Create PDFs with reportlab, review via pdftoppm PNG render loop |
spreadsheets |
skill.md, spreadsheet.md, artifact_tool_spreadsheets_api.md, artifact_tool_spreadsheet_formulas.md + 15 example scripts |
Full spreadsheet creation/editing via openpyxl and a proprietary artifact_tool library |
Notable detail: The spreadsheet skill explicitly instructs the model not to disclose the existence of artifact_tool to users — it's a proprietary internal rendering/recalculation library.
A full PowerPoint generation pipeline:
render_slides.py— main slide rendererpptxgenjs_helpers/— JavaScript helpers forpptxgenjs(SVG, LaTeX, image handling, layout builders)ensure_raster_image.py— converts vector images to raster for embeddingcreate_montage.py— creates image montagesslides_test.py— test suite for slide generation
Part of the CUA (Computer Use Agent) infrastructure. This script merges multiple Chrome Enterprise policy JSON files into a single 000_policy_merge.json, with backups. Used to configure and lock down the Chromium browser instance running inside the container.
Path in container: /openai/project/cua/cua_chrome/cua_chrome/core/policy_merge.py
Live connection info for the IPython kernel at the time of capture:
{
"shell_port": 49539,
"iopub_port": 37165,
"transport": "tcp",
"signature_scheme": "hmac-sha256",
"kernel_name": "python3"
}This confirms the Python execution environment runs as a standard Jupyter kernel communicating over TCP on localhost.
apt_history.log — Full apt installation history. Key packages installed:
- Build tools:
gcc,g++,cmake,ninja-build,make - Media:
ffmpeg,sox,lame,flac,espeak - OCR:
tesseract-ocr+ all language packs (100+ languages) - GIS/data:
gdal,proj,libgeos,libspatialite,libnetcdf - Database clients:
libpq-dev(PostgreSQL),libmariadb-dev,unixodbc - Runtimes:
openjdk-17,php8.2,ruby-full - Desktop:
xvfb,openbox,xfce4(for headless browser/GUI support)
chrome_supervisord.log — Supervisor startup log showing all processes managed at container boot:
| Process | Description |
|---|---|
caas_package_mirror |
Internal package mirror |
policy_merge |
Chrome policy merger (see above) |
xvfb |
Virtual framebuffer (headless display) |
openbox |
Window manager |
xfce4 |
Desktop environment |
mitmproxy |
Man-in-the-middle proxy (network interception) |
x11vnc |
VNC server (remote desktop access) |
novnc_proxy |
noVNC web-based VNC client |
chromium |
Chromium browser instance |
libreoffice |
LibreOffice (for document conversion) |
multikernel_jupyter_tool |
Multi-kernel Jupyter management |
notebook_server |
Jupyter notebook server |
pdf_reader_service |
PDF reading service |
python_tool |
Python execution tool |
terminal_server |
Terminal access server |
rsync_daemon |
File sync daemon |
container_daemon |
Main container controller |
cua_bing_at_home |
CUA Bing integration |
nginx |
Web server / reverse proxy |
| Feature | ChatGPT (OpenAI) | Claude (Anthropic) |
|---|---|---|
| OS | Ubuntu 22.04 | Ubuntu 24.04 |
| Kernel | Linux 5.15 (Azure) | Linux 4.4 (gVisor) |
| Infrastructure | Azure / CAAS | gVisor container |
| Desktop env | Xvfb + Openbox + XFCE4 | None |
| Browser | Chromium (full) | None |
| VNC access | x11vnc + noVNC | None |
| Proxy | mitmproxy | None |
| Skills location | /home/oai/skills/ |
/mnt/skills/ |
| Writable mount | /mnt/data |
/mnt/user-data/outputs |
| Java | OpenJDK 17 | Not present |
| PHP | 8.2 | Not present |
| Ruby | 3.1 | Not present |
This export is provided as-is for educational and research purposes.