›_ Codex Issues
Codex Issues / #19744

Remote SSH connection fails when Node.js is installed via nvm: "No codex found in PATH" and silent app-server launch failure

Open 💬 5 comments Opened Apr 27, 2026 by danohn
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What version of the Codex App are you using (From “About Codex” dialog)?

Version 26.422.30944 (2080)

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Darwin 25.4.0 arm64 arm

What issue are you seeing?

Codex Desktop fails to connect to remote SSH hosts when Node.js (and the codex CLI) is installed via nvm. There are two distinct failures:

  1. "No codex found in PATH" — the PATH probe uses a login+interactive shell (/bin/bash -l -i) but nvm may not initialise, so codex isn't found even though it's installed.
  1. Silent ECONNRESET after the probe succeeds — even when the PATH probe passes (e.g. after symlinking codex to /usr/local/bin), the app-server launch fails silently. Codex Desktop launches the server via a non-interactive SSH command using /usr/bin/env node, which can't find node since nvm isn't loaded. The WebSocket immediately closes with code 1006 (ECONNRESET) and no actionable error is shown in the UI. The actual error (/usr/bin/env: 'node': No such file or directory) is only visible in /tmp/codex-app-server-ssh-ws-v0.log on the remote.

What steps can reproduce the bug?

  1. Install Node.js via nvm on a remote Linux machine
  2. Install the codex CLI via npm (npm install -g @openai/codex)
  3. Add the remote host to ~/.ssh/config on the local macOS machine
  4. Open Codex Desktop and attempt to enable the remote SSH connection
  5. Observe "No codex found in PATH" error
  6. Symlink codex to /usr/local/bin/codex on the remote to get past step 5
  7. Observe persistent connection failure with ECONNRESET — no actionable error shown in the UI

What is the expected behavior?

Codex Desktop should use the same shell environment for both the PATH probe and the app-server launch, so nvm-managed binaries work consistently without manual symlinks. If the launch does fail, the actual error from the remote (e.g. node not found) should be surfaced in the UI rather than a generic connection failure.

Additional information

  • Codex Desktop version: 0.125.0-alpha.3 (macOS)
  • Remote OS: Linux (Debian/Ubuntu), Node.js installed via nvm v24.15.0
  • The workaround requires two symlinks on the remote machine:
ln -sf ~/.nvm/versions/node/$(node --version)/bin/codex /usr/local/bin/codex
ln -sf ~/.nvm/versions/node/$(node --version)/bin/node /usr/local/bin/node
  • The real server-side error is logged to /tmp/codex-app-server-ssh-ws-v0.log on the remote, which is not currently surfaced anywhere in Codex Desktop

View original on GitHub ↗

5 Comments

github-actions[bot] contributor · 4 months ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #19198

Powered by Codex Action

danohn · 4 months ago

The root cause is that /bin/sh on Debian/Ubuntu is dash, not bash. Codex Desktop's
PATH probe runs /bin/sh -lc '...', which sources /etc/profile and ~/.profile but
NOT ~/.bashrc where nvm is initialised. As a result, nvm's bin directory is absent
from the PATH seen by Codex Desktop, even though it is present in regular interactive
SSH sessions.

Dash PATH (what Codex Desktop sees):
/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin

Bash PATH (what a normal SSH session sees):
/root/.local/bin:/root/.nvm/versions/node/v24.15.0/bin:/usr/local/sbin:/usr/local/bin:...

lmxx1234567 · 4 months ago

@danohn Thanks for pointing that out. Here’s my workaround: add the NVM initialization to ~/.profile from ~/.bashrc.

export NVM_DIR="$HOME/.nvm"
if [ -s "$NVM_DIR/nvm.sh" ]; then
  . "$NVM_DIR/nvm.sh"
fi
willdla · 3 months ago
@danohn Thanks for pointing that out. Here’s my workaround: add the NVM initialization to ~/.profile from ~/.bashrc. export NVM_DIR="$HOME/.nvm" if [ -s "$NVM_DIR/nvm.sh" ]; then . "$NVM_DIR/nvm.sh" fi

this works

jiachangliu · 2 months ago

Adding one more data point: I hit the same underlying issue from the ChatGPT iPhone Codex UI connecting to a remote Linux host.

In my interactive shell, Codex worked because nvm was loaded:

command -v codex
codex --version
codex app-server --help >/dev/null && echo app-server-ok

But a clean login-shell simulation could not find codex:

env -i HOME="$HOME" USER="$USER" LOGNAME="$LOGNAME" SHELL=/bin/bash /bin/bash -lc \
'echo "PATH=$PATH"; command -v codex; codex --version; codex app-server --help >/dev/null && echo app-server-ok'

Instead of moving all nvm initialization into ~/.profile, this user-local wrapper also fixed it for me:

NVM_NODE_BIN="$(dirname "$(command -v node)")"
REAL_CODEX="$(command -v codex)"

mkdir -p "$HOME/.local/bin"

cat > "$HOME/.local/bin/codex" <<EOF
#!/usr/bin/env bash
export PATH="$NVM_NODE_BIN:\$PATH"
exec "$REAL_CODEX" "\$@"
EOF

chmod +x "$HOME/.local/bin/codex"

Then the clean-shell check found:

$HOME/.local/bin/codex
codex-cli ...
app-server-ok

After that, ChatGPT iPhone Codex UI was able to connect to the remote Linux host.

This works if ~/.local/bin is already on the login-shell PATH, which is common on Linux via ~/.profile. It avoids sudo, avoids changing /usr/local/bin, and avoids sourcing all of nvm from ~/.profile.

The key difference from moving nvm initialization into ~/.profile is scope: this does not make every login shell load nvm; it only makes codex work when something tries to run codex.