vapi-voice-call

Place outbound phone calls via the Vapi API. Creates the call, polls until completion, and returns structured JSON with transcript.

Installation

bash

# Clone only the vapi-call-cli-tool from rocie-tools
git clone --filter=blob:none --sparse https://github.com/rustykuntz/rocie-tools.git ~/.rocie/tools/vapi-call-cli-tool
cd ~/.rocie/tools/vapi-call-cli-tool
git sparse-checkout set tools/vapi-call-cli-tool

# Install and build
cd tools/vapi-call-cli-tool
npm install
npm run build

The built binary is at ~/.rocie/tools/vapi-call-cli-tool/tools/vapi-call-cli-tool/dist/vapi-call-cli.cjs.

To verify installation:

bash

node ~/.rocie/tools/vapi-call-cli-tool/tools/vapi-call-cli-tool/dist/vapi-call-cli.cjs --help

Required credentials

•VAPI_API_KEY — required. Store via set_config as credentials_tools.vapi_api_key.

Optional credentials

Twilio (for BYO outbound number — only needed when --phone-number-id is not provided and all three are set):

•TWILIO_ACCOUNT_SID — store as credentials_tools.twilio_account_sid
•TWILIO_AUTH_TOKEN — store as credentials_tools.twilio_auth_token
•TWILIO_NUMBER — store as credentials_tools.twilio_number

All credentials are auto-loaded from ~/.rocie/config.yaml.

Usage

bash

VAPI_CLI="$HOME/.rocie/tools/vapi-call-cli-tool/tools/vapi-call-cli-tool/dist/vapi-call-cli.cjs"

# Basic call
node "$VAPI_CLI" \
  --to "+15551234567" \
  --context "Call to confirm booking details and politely close."

# With custom system prompt and force-end after 2 minutes
node "$VAPI_CLI" \
  --to "+15551234567" \
  --context "Ask about order status for order #12345" \
  --system-prompt "You are a customer service agent. Be brief and professional." \
  --hard-end-after 120000

# Context from file
node "$VAPI_CLI" \
  --to "+15551234567" \
  --context-file /tmp/call-brief.txt

# Context from stdin
cat /tmp/call-brief.txt | node "$VAPI_CLI" --to "+15551234567"

Arguments

Arg	Required	Description
`--to`	yes	Phone number in E.164 format (`+` followed by 8-15 digits)
`--context`	yes (or `--context-file` or stdin)	Call objective/context as inline string
`--context-file`	alt	Path to file containing call context
`--system-prompt`	no	Custom system prompt (inline)
`--system-prompt-file`	no	Path to file containing custom system prompt
`--assistant-id`	no	Existing Vapi assistant ID (auto-creates one if omitted)
`--phone-number-id`	no	Vapi phone number ID for outbound
`--timeout`	no	Max wait for completion in ms (default: 300000 / 5min)
`--hard-end-after`	no	Force-end call after N ms (e.g. `120000` for 2min cap)

Output format

Success:

json

{
  "status": "completed",
  "callId": "call_abc123",
  "transcript": [
    {"role": "assistant", "content": "Hi, this is Rocie..."},
    {"role": "user", "content": "Hello, yes I'm here."}
  ]
}

Error:

json

{
  "status": "error",
  "callId": "call_abc123",
  "message": "Call ended with status failed: No answer"
}

Behavior notes

•If no --assistant-id, the CLI auto-creates a temporary Vapi assistant with Rocie's default persona.
•User context from ~/.rocie/context/user.yaml is automatically appended to the system prompt.
•Polls GET /calls/{id} every 2 seconds until terminal state (ended, completed, failed, canceled, no-answer, busy).
•--hard-end-after sends POST /calls/{id}/end to force-terminate long calls.
•Transcript roles are normalized: bot/ai/agent -> assistant, customer/human -> user.

Voice & model defaults

These can be overridden via env vars or ~/.rocie/config.yaml:

Setting	Env var	Default
Voice provider	`VAPI_VOICE_PROVIDER`	`vapi`
Voice ID	`VAPI_VOICE_ID`	`Tara`
Model provider	`VAPI_MODEL_PROVIDER`	`openai`
Model	`VAPI_MODEL`	`gpt-4.1`
First message mode	`VAPI_FIRST_MESSAGE_MODE`	`assistant-speaks-first-with-model-generated-message`
Background sound	`VAPI_BACKGROUND_SOUND`	`office`