michaelxu2288

Elevenlabs Agent Caller — AI skill for Claude Code

AI community

AI agent that makes phone calls for you.

How to install Elevenlabs Agent Caller

This entry records only its repository, not the path inside it, so there is no exact command to give. Open michaelxu2288/elevenlabs-agent-caller and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Elevenlabs Agent Caller does

AI agent that makes phone calls for you. give it a task like ordering from a bakery and it dials the number, talks it through and writes back a transcript + what happened. Twilio for the line, ElevenLabs for the voice and the hearing, Claude as the brain. gets through phone menus, holds and being interrupted. a full fake call runs offline as a test

Alternatives in AI

Prompt Master — A Claude skill that writes the accurate prompts for any AI tool 4.9k ★
Session Kit — Close the terminal windows, not the AI coding sessions 649 ★
Agenvoy — Self-hosted AI agent harness in a single Go binary — writes, sandbox-tests and repairs its own tools, and lets 474 ★

README

elevenlabs agent caller

An AI agent that makes phone calls for you. You write a short task file (who to call, what you want, what you'd accept instead) and it dials the number, has the conversation out loud, hangs up, and leaves a transcript plus a JSON summary of what happened.

The example task is ordering cookies and a loaf of bread from a bakery for pickup. It also works for plain questions to a person.

Twilio is the phone line. The REST API places the call and Media Streams carries the live audio over a WebSocket.
ElevenLabs is the voice (text to speech) and the hearing (speech to text).
Claude is the brain. It runs through the claude CLI you are already logged into, so there is no Anthropic API key.

Everything in between is in this repo: turn detection, barge-in, streaming the reply sentence by sentence, phone-menu key presses, holds, and knowing what the other person actually heard before they cut in.

[00:14] THEM: Sweet Crumb Bakery, this is Maria, how can I help you?
[00:18] AGENT: Hi Maria, I'm an AI assistant calling on behalf of Alex Kim to place a pickup order. Could I get
               a dozen chocolate chip cookies and a loaf of sourdough bread for pickup this Saturday?
[00:27] THEM: Okay, a dozen chocolate chip cookies, got it. Oh, I'm sorry, we're all out of sourdough for the
               weekend. We do have a country white loaf, it's seven fifty. Would that work instead?
[00:37] AGENT (interrupted): Yes, that country white loaf works --
[00:38] THEM: Oh, sorry to cut you off, is this for pickup or delivery?
[00:43] AGENT: Pickup, please.

That excerpt is from the simulated call (see [Dry run and tests](#dry-run-and-tests)).

Where it stands

It has placed about 40 real calls over Twilio from a Mac, to my own phone and to friends who knew it was coming. On those the gap between them finishing a sentence and the agent starting to talk was around 1.7 s (Claude Haiku 4.5, about 0.5 s to first token).
It has n