Codex 1M

macOS patcher for making Codex Desktop run local threads with a 1M-token context window.

What it does:

• starts the Codex Desktop app-server with model_context_window=1000000
• sets model_auto_compact_token_limit=900000
• sets model_auto_compact_token_limit_scope=body_after_prefix
• writes ~/.codex/codex-1m/model_catalog_1m.json
• rewrites every model in that catalog to a visible/effective 1M context window
• rewrites the context usage tooltip to display 1M instead of 1000k
• migrates old local rollout files so reopened threads do not keep stale 258400 context metadata
• removes stale "full context" token markers from old context-window failures
• repacks app.asar in place using a temporary file
• updates Electron ASAR integrity in Info.plist
• ad-hoc signs the patched app
• force-restarts Codex Desktop and kills every Desktop app-server sidecar
• installs the Codex 1M auto-reapply job LaunchAgent for app updates

The LaunchAgent is:

~/Library/LaunchAgents/ai.companion.codex-1m.plist

with launchd label:

ai.companion.codex-1m

Logs go to:

~/Library/Logs/codex-1m/

One-Liners

Apply once and force-restart Codex Desktop:

curl -fsSL https://raw.githubusercontent.com/companion-inc/codex-1m/main/install.sh | bash -s -- apply

Apply now and auto-reapply after app updates:

curl -fsSL https://raw.githubusercontent.com/companion-inc/codex-1m/main/install.sh | bash -s -- install-agent

Check status:

~/.codex/codex-1m/install.sh status

Remove the auto-reapply job:

~/.codex/codex-1m/install.sh uninstall-agent

Notes

The launch args become:

codex app-server --analytics-default-enabled \
  -c model_context_window=1000000 \
  -c model_auto_compact_token_limit=900000 \
  -c model_auto_compact_token_limit_scope=body_after_prefix \
  -c 'model_catalog_json="/Users/you/.codex/codex-1m/model_catalog_1m.json"'

This patches local Codex metadata and UI behavior for all local Desktop threads/models launched through the patched app-server. The model/provider still has to actually accept the requested window.