OpenAI×CallsyWorks with GPT-4, GPT-5, Assistants

Give GPT a voice

Callsy plugs into any OpenAI agent — GPT-4, GPT-5, Assistants API, or the Responses API. Your agent gets 27 tools to place real phone calls, manage contacts, and run voice workflows.

Listen to the voices GPT will use

50+ voices across 40+ languages — click any to hear a sample.

Why GPT + Callsy

Function calling is how GPT reaches the world.
Callsy is how it reaches the phone

OpenAI gave you function calling so agents can browse, search, and execute. Voice was the last channel that didn't have a clean tool surface — building it yourself means juggling Twilio, Deepgram, ElevenLabs, and an agent runtime. Callsy ships it as one function call.

Two integration paths

Native MCP via mcp.callsy.ai/sse, or OpenAI function tools via REST. Either way, 27 tools, zero bridging code you need to maintain.

Works with every surface

Assistants, GPTs, Responses API, Agents SDK, LangChain, LlamaIndex, CrewAI. Anywhere GPT calls functions, it can call Callsy.

Streaming-friendly

make_call returns a call_id instantly. Your agent keeps responding to the user while the phone call runs in the background.

Sub-processor transparency

We list every downstream provider (ElevenLabs, Twilio, Deepgram, etc.) in the Annex. No hidden AI layers, no surprise routing.

Setup

Path A: function tools · Path B: MCP

Most teams start with function tools. Register Callsy as a tool on your OpenAI client:

// Pass this tool into your OpenAI function-calling setup
const tools = [{
  type: "function",
  function: {
    name: "callsy_make_call",
    description: "Place a real phone call via Callsy.",
    parameters: {
      type: "object",
      properties: {
        agent_id: { type: "string" },
        contact_id: { type: "string" },
        context: { type: "object" }
      },
      required: ["agent_id", "contact_id"]
    }
  }
}];

// Then forward the call to Callsy's REST API
await fetch("https://api.callsy.ai/v1/calls", {
  method: "POST",
  headers: { "Authorization": `Bearer ${process.env.CALLSY_KEY}` },
  body: JSON.stringify(args)
});

Then just prompt GPT naturally:

Call Sarah about her abandoned cart worth $142.
Offer free shipping. Use Emma's warm US English voice.
Report back with the transcript and outcome.

Workflows GPT handles end-to-end

One prompt · Every step

Outbound sales

Take this CSV of 50 leads, score them, call the top 10, and book demos for the qualified ones.

Cart recovery

Pull cart_abandoned contacts from the last 48h. Call them with personalized 10%-off offers.

Customer support callback

For every support ticket flagged urgent, call the customer within 10 minutes.

Renewal outreach

Call customers whose subscription expires this month. Offer the loyalty discount.

Frequently asked

How do I connect Callsy to OpenAI?

Two paths. (1) If your OpenAI client supports MCP natively, point it at https://mcp.callsy.ai/sse. (2) Otherwise, register Callsy's REST endpoints as OpenAI function tools — we publish a ready-to-use tool schema.

Does this work with GPT-4, GPT-5, and Assistants?

Yes. Any OpenAI model with function calling can call Callsy. That covers GPT-4 (all variants), GPT-5, Assistants API, and the Responses API.

Can I use this with Agents SDK?

Yes. Callsy exposes the same 27 tools whether your agent framework is the raw OpenAI Assistants API, the Agents SDK, LangChain, LlamaIndex, or CrewAI.

What's the latency?

Tool-call round-trips are typically 200–800ms. The actual call happens asynchronously — make_call returns a call_id immediately, then your agent polls get_call for the outcome.

Will my conversation data train OpenAI?

That depends on your OpenAI account's data settings, not Callsy. Separately, Callsy itself does not train foundation models on your call data.

Plug Callsy into GPT

Free tier · 10 calls · no credit card · 5-minute setup.

Start with OpenAI