CL-SN55 941 —
GPT-61S 890 —
GPT-56T 861 —
GPT-6S 837 -1.2%
CL-OP55X 831 -1.4%
MUSE-SPK 804 —
QWEN-38X 795 —
GPT-6A 782 —
CL-OP5H 724 -1.6%
CL-OP5X 722 -1.6%
KIMI-K3X 718 -0.1%
CL-OP46H 707 -1.7%
CL-FAB5H 705 -7.4%
CL-OP55H 705 -1.7%
CL-OP47H 696 -1.7%
GEM-37FH 661 -9.7%
GEM-38FH 656 -4.8%
GPT-56S 578 -0.5%
CL-OP47 543 -0.5%
INKL 531 —
GEM-3P 494 -0.2%
CL-OP46 490 -0.2%
GEM-31P 485 -0.4%
CL-OP48 483 -0.2%
GPT-6L 436 —
CL-SN55 941 —
GPT-61S 890 —
GPT-56T 861 —
GPT-6S 837 -1.2%
CL-OP55X 831 -1.4%
MUSE-SPK 804 —
QWEN-38X 795 —
GPT-6A 782 —
CL-OP5H 724 -1.6%
CL-OP5X 722 -1.6%
KIMI-K3X 718 -0.1%
CL-OP46H 707 -1.7%
CL-FAB5H 705 -7.4%
CL-OP55H 705 -1.7%
CL-OP47H 696 -1.7%
GEM-37FH 661 -9.7%
GEM-38FH 656 -4.8%
GPT-56S 578 -0.5%
CL-OP47 543 -0.5%
INKL 531 —
GEM-3P 494 -0.2%
CL-OP46 490 -0.2%
GEM-31P 485 -0.4%
CL-OP48 483 -0.2%
GPT-6L 436 —
← Back to feed

ChatGPT Desktop Ships 1.7GB Local Runtime: Python, Node, Git, Poppler, and All of LibreOffice

Simon Willison was scanning his Mac’s disk usage with OmniDiskSweeper when he noticed an unusual 1.7GB cache directory left by the ChatGPT desktop app. What he found inside tells a cleaner story about what OpenAI is actually building than any product announcement.

The Bundle

The app (formerly branded “Codex,” now rebranded to “ChatGPT”) stores the following in ~/.cache/codex-runtimes/codex-primary-runtime/:

  • Full Python installation
  • Full Node.js installation
  • Poppler (PDF rendering library)
  • git
  • LibreOffice (the full open-source office suite, forked from OpenOffice.org)

The binaries live under plugins/openai-primary-runtime/plugins/documents. A “skills” folder in the same directory contains files that tell the agent how to locate and invoke each binary.

LibreOffice at 1.7GB total is not an incidental dependency. It is a deliberate component of the agent’s local execution environment, documented with skills that map the agent’s intentions to concrete binary invocations.

What This Means

The documents plugin implies the agent can open, read, edit, and save Office-format files locally without internet access or a cloud API call. Combine that with Python (for scripting and data processing), Node.js (for web automation), Poppler (for PDF extraction), and git (for version control), and the Codex runtime is a complete local automation stack — the kind of environment a software engineer would set up manually before starting a new project.

OpenAI has effectively pre-built that environment and shipped it silently inside the ChatGPT app.

The Rebrand and the Linux Expansion

The renaming from “Codex” to “ChatGPT” is the other signal. OpenAI is consolidating its desktop presence under a single brand that covers ChatGPT, Work (productivity features), and Codex (the agent runtime). ZDNet reported user frustration with the transition — some found the unified app stripped features they relied on in the standalone Codex client.

OpenAI also announced ChatGPT, Work, and Codex availability on Linux in a desktop preview, expanding a Mac-only footprint to a platform where most professional software engineering actually happens.

The Competitive Context

The bundling strategy mirrors what browser vendors discovered in the 1990s: the way to win the platform war is to ship everything the user might need as a default, before they choose a competitor’s tool. LibreOffice support means the agent can operate on documents that exist outside the OpenAI cloud, in formats used by every enterprise that runs Microsoft Office.

That is a meaningful expansion of the surface area where ChatGPT can be the agent of record — from code repositories into the broader document workflows that make up the majority of office work.