Build a Voice Agent with VAPI Functions and Custom Tools (2026)

TL;DR — VAPI's "Custom Functions" feature was deprecated in 2026 in favor of the Tools API. Tools are reusable across assistants, support async webhooks, and have a clean HMAC-signed contract. This is the only correct way to extend a VAPI agent today.

What you'll build

A VAPI assistant that calls a book_appointment tool against your Next.js webhook, validates the HMAC signature, books the slot in Postgres, and returns a confirmation the assistant speaks aloud.

Architecture

flowchart LR
  CL[Caller] --> VP[VAPI assistant]
  VP -- function_call --> TL[Tool: book_appointment]
  TL -- HMAC POST --> WH[Your /api/vapi/book webhook]
  WH -- result JSON --> VP --> CL

Step 1 — Create the tool in dashboard

In dash.vapi.ai → Tools → Create. Pick Custom Tool and fill:

• Name: book_appointment
• Server URL: https://yourhost.com/api/vapi/book
• Description: "Book a slot once the user confirms an ISO-8601 date+time."
• Parameters JSON:

```json { "type": "object", "properties": { "iso": { "type": "string", "description": "ISO-8601 datetime" }, "name": { "type": "string" } }, "required": ["iso", "name"] } ```

Step 2 — Attach to assistant

In Assistants → Edit → Tools, toggle book_appointment on. The system prompt should mention "Use book_appointment once the caller confirms."

Step 3 — Webhook handler (Next.js)

```ts // app/api/vapi/book/route.ts import crypto from "crypto"; import { db } from "@/lib/db";

export async function POST(req: Request) { const raw = await req.text(); const sig = req.headers.get("x-vapi-signature") ?? ""; const expected = crypto.createHmac("sha256", process.env.VAPI_WEBHOOK_SECRET!) .update(raw).digest("hex"); if (!crypto.timingSafeEqual(Buffer.from(sig), Buffer.from(expected))) { return new Response("bad sig", { status: 401 }); } const body = JSON.parse(raw); const call = body.message.toolCallList[0]; const { iso, name } = call.function.arguments; await db.appt.create({ data: { iso, name } }); return Response.json({ results: [{ toolCallId: call.id, result: Booked ${name} for ${iso}., }], }); } ```

Step 4 — Programmatic create

```ts const r = await fetch("https://api.vapi.ai/assistant", { method: "POST", headers: { Authorization: Bearer ${process.env.VAPI_API_KEY}, "Content-Type": "application/json" }, body: JSON.stringify({ name: "Clinic Concierge", model: { provider: "openai", model: "gpt-4o", toolIds: ["tool_book_appointment"] }, voice: { provider: "11labs", voiceId: "rachel" }, transcriber: { provider: "deepgram", model: "nova-3" }, firstMessage: "Hi — Sunrise Clinic, how can I help?", }), }); ```

Step 5 — Async tools

For long-running work (>5s), set async: true on the tool. VAPI keeps the user engaged with a fill phrase ("One sec while I check that...") and waits up to 60s for your webhook to POST a result back via the /call/:id/control endpoint.

Step 6 — Test with the simulator

vapi-cli simulate --assistant <id> --transcript 'I want 3pm tomorrow' exercises the tool path without spending phone minutes.

Pitfalls

• toolCallId mismatch: Your response results[].toolCallId MUST match the request call.id — otherwise VAPI ignores the result.
• HMAC vs server JWT: Some old docs reference Bearer tokens — Tools use HMAC by default in 2026.
• Idempotency: VAPI retries on 5xx — return 200 even on duplicate, with the same result body.
• Tool description quality: One job per tool, plain-English description; the LLM picks tools by description, not name.

How CallSphere does this

CallSphere ships 37 agents · 90+ tools · 115+ DB tables · 6 verticals, including a VAPI tier for low-volume franchise customers. $149/$499/$1,499 · 14-day trial · 22% affiliate.

FAQ

Default vs custom tools? VAPI ships defaults like endCall, transferCall, dtmf — use them for control flow and reserve custom tools for business logic.

Latency? Tool execution adds your webhook RTT; keep p95 under 800ms or VAPI fills with a "still checking" line.

Streaming responses? Tools are request/response only — for streaming use a Custom LLM URL instead (separate feature).

Multi-step workflows? Use VAPI Workflows (visual graph) or chain tool calls in your prompt.

Sources

• VAPI Docs - Custom Tools - https://docs.vapi.ai/tools/custom-tools
• VAPI Docs - Default Tools - https://docs.vapi.ai/tools/default-tools
• VAPI Docs - Assistants Quickstart - https://docs.vapi.ai/assistants/quickstart
• ChatBotKit - Build Custom Voice Agents with Vapi - https://chatbotkit.com/tutorials/how-to-build-custom-voice-agents-with-vapi-and-chatbotkit