Getting started
Connect Codex — one line from a terminal
Codex is OpenAI's coding agent. There are two ways to give it Erawan, and they are good at different things — this is both of them, and how to tell which one you want.
The connector
When you add a --url rather than a command, Codex uses the HTTP transport on its own — there is no transport flag to get right. The first time it actually calls Erawan, a browser opens so you can approve the connection; nothing is pasted and there is no key to keep.
That confirms it is registered. Then say what you want built, and ask for it to be put online.
The other way, and why it is often better
That installs the erawan command, and — if you already use Codex — leaves a short pointer in ~/.codex/AGENTS.md, the file Codex reads in every session. It does not dump the whole playbook there: that file is loaded for every project you ever open, and a hundred lines about deploying would be paid for by every session that never deploys. The pointer names the command, the exit codes, and where the real instructions are.
The difference that matters: the connector gives Codex tools, and the CLI gives it a procedure — check the credential, assess, deploy, poll, read the build log, fix, same name again. When something fails, the second is what turns "it did not work" into a fix.
Which to use
Both, if you can. If Codex is running somewhere with no shell — a sandbox, a hosted environment — the connector is the one that works. If it has a terminal, install the CLI too: it is what makes the agent able to read its own build log.
What to say once it is connected
You do not have to name Erawan or explain the steps. If either route is in place the agent already knows how, and what comes back is a link anyone can open.