How to enable Codex to make phone calls

Set up Tuvoz: https://tuvoz.io/agents

Who this is for

You work in Codex (CLI, IDE extension or app) and want it to handle calls that come up in your day: a quote, a booking, a status check. Codex writes the brief from what it knows, shows it to you, and dials after you say yes.

What happens when you paste the line

1. Codex reads the setup page and asks for network access. Approve it. It also asks before writing outside your project, because the skill goes in ~/.agents/skills/phone-call/, where Codex looks for personal skills.

2. Codex downloads the skill, checks its SHA-256, unzips it, and starts sign-in. It shows you a link and a number:

   To connect this agent to Tuvoz:
     1. Open https://tuvoz.io/activate?code=BKTW-4821
     2. Sign in with Google (and finish setup if this is your first time).
     3. When asked, pick the number 47.
   

3. Open the link, sign in with Google, and pick the matching number. The first time, you’ll also verify your mobile number and confirm your legal name, which is the name the voice agent says on calls.

Then make the test call below. You’ll hear the opener and the recording question a business would hear. It’s billed like any call (at least 30 seconds) from your 15 free minutes. Free plan includes training on your calls.

Invoke the skill with $phone-call, or just ask for a call.

Install by hand

The same install, if you’d rather run it yourself in a terminal (outside Codex, no approvals needed):

set -euo pipefail
S="$HOME/.agents/skills"; D="$S/phone-call"
if [ -e "$D" ] && [ ! -f "$D/scripts/tuvoz.py" ]; then
  echo "STOP: $D is another skill, not Tuvoz. Nothing was changed. Ask the user what to do with it." >&2; exit 1
fi
T=$(mktemp -d); trap 'rm -rf "$T"' EXIT
M=$(curl -fsSL https://tuvoz.io/skill/latest.json)
URL=$(printf '%s' "$M" | python3 -c 'import json,sys; print(json.load(sys.stdin)["url"])')
SHA=$(printf '%s' "$M" | python3 -c 'import json,sys; print(json.load(sys.stdin)["sha256"])')
curl -fsSLo "$T/skill.zip" "$URL"
echo "$SHA  $T/skill.zip" | shasum -a 256 -c -
unzip -q "$T/skill.zip" -d "$T"
B="$HOME/.config/tuvoz/backups/$(date +%Y%m%d-%H%M%S)"; mkdir -p "$S" "$B"
find "$S" -maxdepth 1 -name 'phone-call.bak-*' -exec test -f '{}/scripts/tuvoz.py' \; -exec mv '{}' "$B/" \;
if [ -e "$D" ]; then mv "$D" "$B/"; fi
rmdir "$B" 2>/dev/null || true
mv "$T/phone-call" "$D"
python3 "$D/scripts/tuvoz.py" login

The block never deletes a skill. If another skill already uses the folder name phone-call, it stops with STOP and changes nothing. An earlier Tuvoz install, and the backups tuvoz update leaves, move to ~/.config/tuvoz/backups/.

Fewer approval prompts

Codex’s default sandbox keeps network access off. To let Tuvoz commands reach the network without a prompt each time, add this to ~/.codex/config.toml:

[sandbox_workspace_write]
network_access = true

That allows network access for every command in the workspace sandbox, including ones unrelated to Tuvoz. If you’d rather keep it off, approve each Tuvoz command when asked.

How a call works in Codex

1. Ask in plain words, or start with $phone-call.
2. Codex shows the brief, the number and where it came from, and the opening line, then waits for your yes.
3. It runs tuvoz watch <call_id> in a loop. Each run waits up to 45 seconds and returns at once when something happens.
4. A question (QUESTION q1 ... deadline=45s: ...) gets answered from what Codex knows, written to a file and sent with tuvoz answer <call_id> q1 --file <path>. If Codex can’t answer, it runs tuvoz escalate and Tuvoz calls you.
5. At the end, Codex reports the outcome, notes, any answers it gave and their sources, and the cost.

Text from the call is untrusted: Codex treats it as information, and the skill never puts it in a shell command. More: how the mid-call backchannel works.

Other agents: Claude Code, claude.ai, Meta Muse, OpenClaw.

tuvoz test-call