wavekat
voice docs blog
← back to blog

Let Your AI Assistant Place Real Phone Calls

· Eason Guo
#voice-ai #automation #ai-agents

WaveKat Voice now ships with a command-line tool, so a program you trust — including an AI assistant like Claude — can place and manage real phone calls for you. Ask your assistant to “call the dentist and wait until someone picks up,” and it dials through the app you already have open, follows the call, and tells you how it went. It’s built into the app today on Mac and Linux, and it’s off until you switch it on.

This is the next step toward the thing we keep coming back to: giving every small business the voice of a big one. A big company has a switchboard and software that drives it. Now your computer — and the assistant running on it — can be that switchboard.

What it actually does

WaveKat Voice has always had a phone running quietly in the background: it registers with your SIP provider and handles calls. What’s new is a second way to drive it — a command called wavekat-voice that talks to the running app.

To be precise about the boundary, because it matters:

So the assistant is the hands on the dialpad, not a voice on the line. That’s a deliberate, honest line — and for the everyday “get me through to a human” chores, it’s most of what you actually want.

WaveKat Voice on Ubuntu — a call the assistant placed, in progress, with a live transcript alongside.

There’s nothing to install

The wavekat-voice command is the same program that runs the app — it’s already on your disk the moment you install WaveKat Voice. There’s no second download, no separate package, no version that can drift out of sync with the app.

It is off by default. While automation is on, any program you run on your computer can place calls through your account — and calls may cost money — so we leave that decision to you. Turn it on in Settings → Automation, where there’s also a one-click button to add wavekat-voice to your PATH so any terminal can find it.

WaveKat Voice on Ubuntu — Settings → Automation, with command-line access turned on and the Install command-line tool button.

Connect an AI assistant in one click

The fastest path is the Settings → Automation page itself. It looks for AI assistants you already have installed and offers a Connect button for each. Today that covers:

AssistantHow it connects
Claude Desktop, Cursor, WindsurfThrough an MCP server bundled in the app
Claude Code, Codex, GeminiThrough a managed note in their instructions file

One click wires it up — nothing to copy or paste. After that, you just ask the assistant to make a call. Two things worth knowing: some assistants need a full restart (quit and reopen) to pick up the new tools, and the connection keeps itself current — when WaveKat Voice updates in the background, any assistant you’ve connected is quietly kept in sync, so you never have to reconnect.

WaveKat Voice on Ubuntu — connecting AI assistants like Claude and Cursor, each with a one-click Connect button.

What it looks like from a terminal

Every command takes --json for machine-readable output, which is what makes it comfortable for an assistant to drive. A few examples:

# Is the app running, and which accounts are connected?
wavekat-voice status

# Place a call and wait — the exit code says how it went.
wavekat-voice call +14155550123 --wait
echo "result: $?"

# Find a call that's happening right now, then hang it up.
wavekat-voice call list --json | jq -r '.[0].id' | xargs wavekat-voice call hangup

The --wait exit code is the contract a script (or an assistant) branches on: 0 answered then ended normally, 2 busy or declined, 3 failed or dropped, 4 no answer. No output parsing required to know what happened.

Commands are grouped by what they act on — call for placing and handling calls, recording for saved audio, log for the activity log — with status, accounts, and a live events stream at the top level. Run wavekat-voice call --help to see the full set.

Why we built it this way

A few choices we’re happy with:

Frequently asked questions

Can an AI assistant place phone calls with WaveKat Voice?

Yes. With automation enabled in WaveKat Voice (Settings → Automation), an AI assistant like Claude can place, follow, and end real phone calls through the app’s command-line tool or its MCP server. The assistant drives the call; you speak on it.

Does the AI talk on the call instead of me?

No. WaveKat Voice routes call audio through your computer’s microphone and speakers — you do the talking. The assistant handles dialing, waiting for an answer, sending menu tones, and hanging up.

Do I need to install anything extra to use the command line?

No. The wavekat-voice command ships inside the WaveKat Voice app, so it’s already on your computer. You only need to turn on automation in Settings → Automation, and optionally click “Install command-line tool” to add it to your PATH.

Is it safe to leave automation on?

Leave it off unless you’re using it. While automation is on, any program you run on your computer can place calls through your account, which may cost money. It’s off by default for that reason, and you can switch it off again at any time.

Which assistants can connect in one click?

Claude Desktop, Claude Code, Cursor, Codex, Gemini, and Windsurf today — via a bundled MCP server for the desktop assistants and a managed instructions note for the command-line ones.

Which platforms support this?

WaveKat Voice runs on Mac and Linux today, with Windows coming when there’s demand. The command-line tool and assistant integrations are available on both supported platforms.

Try it

Download WaveKat Voice, open Settings → Automation, and connect your assistant. The full command reference — every command, its JSON output, and the exit codes — lives in the automation docs.

We’re just getting started here. Driving calls is the foundation; an assistant that can also hold the conversation is where this goes next.

← back to blog