Skip to main content
Beta Codex is OpenAI’s coding agent, in the terminal and as a desktop app. It talks to the Responses API, which Lyceum serves in beta, so you add Lyceum as a model provider and keep the rest of your setup.

​
Configure

Add a Lyceum provider to ~/.codex/config.toml: web_search = "disabled" turns off Codex’s built-in web search, which doesn’t work on Lyceum yet. See Web search for a workaround. Export your key, then start Codex in your project:
Already use Codex with OpenAI? Put the same lines in ~/.codex/lyceum.config.toml instead, and start Codex with codex --profile lyceum when you want Lyceum.
Codex’s built-in web search doesn’t work on Lyceum yet: it runs on OpenAI’s servers. Until we serve it, add a search tool through MCP instead. Exa and Brave both include free credits.
Exa’s hosted server needs nothing installed. Add it to ~/.codex/config.toml:
~/.codex/config.toml
Codex calls the search tool whenever a task needs the web. Search queries go to the provider you choose, not to Lyceum.

​
Verify

Run one task without the interactive view: To let Codex edit files in a one-off run, add -s workspace-write. Continue the last session with codex exec resume --last "<your next prompt>".

​
Notes

  • Desktop app. The Codex app reads the same config.toml. Pick the Codex workspace in the switcher at the top left, then start a project.
  • Models. Kimi K3 and DeepSeek V4.1 Flash work well for coding in Codex. Take model IDs from the models page.
  • Reasoning effort. Kimi K3 accepts low, high and max. If you set model_reasoning_effort, use one of those.
  • Model notice. Codex says it has no metadata for the model and uses defaults. You can ignore it: tools and file edits work.
  • History. We don’t store responses. Codex sends the whole conversation with every turn, so resumed sessions work as usual.
  • Beta. Requests and streaming over HTTP work today, and we’re adding more of the Responses API. Tell us at support@lyceum.technology what you need next.