The ChatGPT/Codex app bundles a full copy of LibreOffice
TL;DR Highlight
Poking through ~/.cache revealed the Codex app ships 1.7GB of runtimes including LibreOffice
Who Should Read
A backend developer figuring out how to let an agent read and write office files like docx and xlsx
Core Mechanics
- Codex cache folder was 1.7GB
- Simon Willison spotted it while browsing ~/.cache with OmniDiskSweeper.
- The folder is named codex-primary-runtime.
- The OpenAI Codex desktop app has since been rebranded to ChatGPT.
- Full Python and Node.js ship inside
- A full Python installation and a full Node.js installation are included.
- Native binaries for Poppler, git and LibreOffice are bundled too.
- LibreOffice forked from OpenOffice.org in 2010
- The post presents it as one of the bundled components, an open source office suite.
- A documents plugin holds the usage guide
- The plugins/openai-primary-runtime/plugins/documents folder contains skills.
- Those skills tell Codex how to find and use the binaries.
Community Reactions
- Commenter says LibreOffice is the answer for old xls files
- One commenter bundles LibreOffice in their own app as well.
- They found nothing else they could trust to read old xls files.
- They now use it for all document-related work.
- Many guessed it is for format conversion
- Some suggested it runs headless to convert file formats back and forth.
- One commenter said Codex's reasoning snippets often mention LibreOffice when editing Word files.
- Another suggested it supports the new Work product that produces docx and pdf.
- Commenters pushed back on bundling
- One asked why it does not first look for binaries the user already has.
- Another argued bundling has no security benefit over downloading hash-locked components on demand.
- A commenter questioned license compliance
- They could not find LibreOffice in the app's open source licenses section and asked if it violates MPL 2.0.
- The original post does not address this.
How to Apply
- Check your own cache folder sizes today
- Open ~/.cache with OmniDiskSweeper or du and see how big agent app folders are.
- Look inside for bundled runtimes and skills folders.
- Write a skill for office file handling
- Document where the binaries live and how the agent should call them.
- Add example commands for each task such as conversion or rendering.
- Decide on bundling LibreOffice at design time
- Consider it if compatibility with formats like old xls matters.
- If you bundle it, check app size and license notice obligations.
Try It
# Paste into Claude Code or Codex (a light version of checking the finding)
Find the 10 largest folders under ~/.cache in my home directory.
If any look like they were created by an AI agent app, report:
1. Folder size
2. The runtimes (Python, Node.js, etc.) and native binaries inside
3. If there is a skills or plugins folder, which tools its docs explain how to use
Read only. Do not delete or modify any files.
# Next to try: ask it to read the skills docs in {{folder path}} and summarize the office file conversion stepsTerminology
Related Papers
Same Simulation, Different Evidence: Where Retrieved Tool Output Costs LLM Agents Accuracy
Saving long tool output to files cut GPT-5.1 accuracy by 11.3 points; naming the authoritative file restored it.
Fake News, Real Crashes: Information Injection Attacks on LLM Financial Trading Agents
Fake headlines hijacked LLM trading agents 62% of the time versus 14% for prompt injection.
Harness Engineering 101: The LLM is the brain, the harness is the body
An agent is a loop managing a JSON array; 16 chapters build one from scratch
Harness2 by Google Research – Recursive Agent Harnessing
Recursively refining the harness with fixed model weights raised benchmark scores by up to 11.8 points.
Ousted Savannah police officer gave ICE agent husband access to Flock system
A police officer shared her Flock login with her ICE-agent husband; an audit found 983 searches and she was fired
Tmux and Claude Code: Sessions, Send-Keys, and the Ceiling
Running Claude Code agents in tmux works well, but limits appear around four agents.