llm-openai-codex 0.4.0

Access OpenAI models via a Codex subscription

llm-openai-codex

LLM plugin for accessing ChatGPT/Codex-backed OpenAI models through the Responses API.

Fork of Simon Willison's llm-openai-via-codex with a codex/ model prefix, plugin-owned auth, llm codex commands, verbosity, and OpenAI's server-side web_search tool.

Changes from the original plugin

• Package renamed to llm-openai-codex.
• Model prefix changed to codex/.
• Plugin-owned auth is stored in LLM's user config directory as auth-codex.json, with borrowed fallback to ${CODEX_HOME:-~/.codex}/auth.json.
• llm codex auth commands manage login, import, status, refresh, and logout.
• Explicit verbosity option maps to Responses API text.verbosity.
• Extra Responses API options are forwarded when LLM accepts them.
• Missing account_id values are derived from OAuth JWT claims when possible.
• Added llm codex usage showing current Codex plan usage.
• Registers known hidden/fallback Codex models such as gpt-5.3-codex-spark.
• Added OpenAI's server-side web_search tool via -o web_search 1.

Installation

Install this plugin in the same environment as LLM.

llm install llm-openai-codex

Usage

Authenticate the plugin:

llm codex login [--device-code]

If you already use Codex CLI, the plugin can fall back to ${CODEX_HOME:-~/.codex}/auth.json when auth-codex.json does not exist. You can also copy those tokens into plugin-owned auth with llm codex import.

List available Codex-backed models:

llm models -q codex

Models are discovered from the Codex API, plus known hidden/fallback slugs such as gpt-5.3-codex-spark. Some listed fallback models may still fail at request time if your plan lacks access.

The discovered list is cached for 24 hours in codex_models.json in LLM's user config directory; delete that file to force a re-fetch.

Run a prompt:

llm -m codex/gpt-5.3-codex-spark "Hello"

Use Responses API verbosity:

llm -m codex/gpt-5.3-codex-spark -o verbosity low "Summarize this"

Image attachments (-a image.png) are sent at low detail to limit token use.

Web search

Enable OpenAI's server-side web_search tool (the search runs on OpenAI's servers, not locally):

llm -m codex/gpt-5.5 -o web_search 1 "What happened in AI news today?"
llm -m codex/gpt-5.5 -o web_search 1 -o web_search_live 1 "Latest stable Python release?"
llm -m codex/gpt-5.5 -o web_search 1 -o web_search_context_size high "Summarize recent LLM benchmarks"

web_search_live requests live internet access instead of the cached index; web_search_context_size is one of low, medium, high. Availability depends on your plan and model.

To enable web search by default for a given model, use LLM's per-model default options:

llm models options set codex/gpt-5.5 web_search 1

Auth commands

llm codex login
llm codex login --device-code
llm codex status
llm codex refresh
llm codex refresh --borrowed
llm codex usage
llm codex import
llm codex logout

Auth source order:

1. Plugin-owned auth-codex.json
2. Borrowed Codex CLI auth at ${CODEX_HOME:-~/.codex}/auth.json

llm codex import copies ChatGPT OAuth tokens from Codex CLI auth into plugin-owned auth-codex.json. It refuses to overwrite an existing auth-codex.json; run llm codex logout first if you want to replace plugin-owned auth.

Refresh behavior

• Plugin-owned auth refreshes lazily and on demand via llm codex refresh.
• Borrowed Codex CLI auth never refreshes automatically. When it expires, pick one:
  • llm codex refresh --borrowed — refresh the shared file in place. Rotates the shared refresh_token; restart any running Codex CLI session afterwards.
  • Run Codex CLI itself to refresh.
  • llm codex import to promote it to plugin-owned auth.

llm codex status shows which auth source is active.

When authentication is missing or expired, run llm codex login [--device-code] or llm codex import.

Development

uv run pytest
uv run llm plugins
uv run llm codex status

Releasing

1. Bump version in pyproject.toml (PEP 440, no v prefix, e.g. 0.2.4).
2. Commit, then tag with a v prefix and push:

   git tag v0.2.4 && git push origin v0.2.4
   

3. Publish a GitHub Release for that tag (gh release create v0.2.4 --generate-notes). This triggers .github/workflows/publish.yml to test and publish to PyPI.