Keep Codex’s API keys out of .env files and chats.

OpenAI’s coding agent, in the terminal and the IDE. With Vando, your keys stay on your phone. When Codex needs one, your phone shows who is asking and why, and you give it just what it asked for, for one run or a set time, or you deny it with one tap.

The Codex guide, step by step

Checked against Codex’s own documentation on 2026-11-10; the guide links its sources.

Set up Codex in four steps.

  1. 1 · Install vando and connect it to your phoneIn your own terminal, not through Codex. vando connect shows a code; open Vando on your phone, tap Connect an AI, enter it and tap Connect. The code works once, for 10 minutes.
    $ npm i -g @getvando/cli --registry=https://registry.npmjs.org --@getvando:registry=https://registry.npmjs.org
    $ vando connect
  2. 2 · Tell it how to askPut the note in AGENTS.md in the repository root. Codex reads one from each folder between the root and where you are, so a note higher up covers the folders below.
  3. 3 · Let vando run reach your phoneCodex runs commands in a sandbox whose network is off in the normal workspace-write mode, so vando run can’t reach your phone until you allow it. The cleanest way is a rule that lets vando run run outside the sandbox (Codex marks rules as experimental), in ~/.codex/rules/default.rules. decision = "prompt" asks you in Codex each time and then your phone asks; "allow" skips Codex’s prompt, and the phone is still the gate. The other two ways: network_access = true under [sandbox_workspace_write] in ~/.codex/config.toml, or Codex’s network proxy with only Vando’s server allowed.
    prefix_rule(pattern = ["vando", "run"], decision = "prompt",
                justification = "Vando asks the phone itself")
  4. 4 · The MCP tools, if you want themOptional. Add the server to ~/.codex/config.toml, shared by the CLI, the IDE extension and the desktop app, or run codex mcp add vando -- vando mcp. Each tool call times out after one minute by default; the longer limit gives a phone approval room.
    [mcp_servers.vando]
    command = "vando"
    args = ["mcp"]
    tool_timeout_sec = 90

Use Vando in every project.

Codex reads ~/.codex/AGENTS.md in every project. Read once per run (restart Codex after editing), before the project’s AGENTS.md files; the closer file wins where they disagree. AGENTS.override.md in the same folder replaces it. (Codex’s AGENTS.md docs.)

Put this note there, so Codex gets secrets through Vando everywhere, suggests vando add in a new project, and suggests moving a project’s .env files into Vando when it finds them:

## Secrets (Vando), in every project
- Never open, print, copy or edit `.env*` files or secret values, and never put a secret in chat.
- Run anything that needs a secret as `vando run --secret NAME -- <command>` (repeat `--secret` for more than one).
- In a new project, keep secrets in Vando from the start: suggest `vando add NAME` for each one, then `vando run` to use it.
- In a project that has `.env` files with values in them, suggest `vando init`: it moves them into Vando with one Save on the owner's phone, so the values stop living in plain files that you and other tools can read.
- Exit 75 means it is waiting for the owner's phone: say "Approve NAME on your phone, then I'll run it again", do other work, then run the same command again. After two waits, stop and tell the owner to open Vando on their phone (the request is in the Inbox).
- Exit 77 means the owner said no or the access ended: don't retry; ask the owner.

An example: A refund script with your Stripe key, from inside the sandbox.

You
Ask Codex to run the refund script.
Codex runs
vando run --secret STRIPE_KEY -- node scripts/refund.js
Codex shows
Its own prompt for the command (if your rule says prompt), then “Waiting for your phone”.
Your phone shows
“Codex · MacBook wants to use STRIPE_KEY”. Give once, with your fingerprint; the script runs with the key in its environment only.

In the cloud, or on another computer.

Codex cloud runs tasks in its own environments, with an install script and a list of allowed domains. You can install vando there and allow Vando’s server, but a cloud task has nobody to read the connect code and each task starts clean. Never store a Vando connection key in Codex’s secrets: that would bypass your phone. Use Vando with the Codex CLI or IDE on a computer you sit at.

Good to know with Codex.

Questions.

Why does Codex need a rule for Vando?

Codex’s default sandbox has no network, so a command that has to ask your phone is blocked until you allow it. A prefix rule for vando run is the narrowest way; opening the network for workspace-write or allowing only Vando’s server through the proxy are the other two.

Does Codex read AGENTS.md?

Yes. Codex reads AGENTS.md from ~/.codex and from each folder between the repository root and where you are, with the closer file winning where they disagree.

Can Codex cloud use Vando?

It can install vando and reach Vando’s server if you allow the domain, but nobody is there to read the connect code and each task starts clean. Keep Vando for the Codex CLI or IDE on a computer you sit at, and never put a connection key in Codex’s secrets.

What Vando is, in short.

One app on your phone for your passwords, 2FA codes, recovery kits, files and developer secrets. Each thing gets its own lock: Easy, Just me, Two devices or Fort Knox, from a quick tap up to your fingerprint plus a security key. Your laptop and your AI tools hold no keys; when one needs a login or a key, your phone shows who is asking and for what, and you say yes or no there. Every yes is signed on your phone and kept on record.

How the locks work → · The detail for developers → · Every AI tool you can set up with Vando → · Get the app →