Enterprise DNA
Directories / Use Cases / Build a Voice Agent

Use case

Build a Voice Agent

A phone-answering AI that qualifies inbound callers, handles the most common questions without a human, and escalates the right calls with a full context handoff.

Your front-desk phone rings after hours, during a busy period, or faster than your team can answer. Without a voice agent, those calls go to voicemail and the lead goes cold. The agents category has two hard problems: first, the real-time audio pipeline (STT, LLM, TTS all need to stay under about 800ms or the conversation feels broken), and second, the qualification logic itself (which questions to ask, when to escalate, what to write back to the CRM). Founders, ops leads, and solo-service businesses are the builders here, typically trying to replace a phone-answering service or extend coverage beyond business hours without hiring.

The stack

Each pick is a real entry on the index. Click any one for the full detail page.

  1. 1
    A Agents Call orchestration

    Vapi

    by Vapi AI

    Why this: Vapi handles the telephony plumbing, provider-agnostic STT and TTS wiring, tool calling, and call observability in one platform. For a qualification-focused voice agent, the webhook-and-tools model is the right shape: the agent calls a tool to check a CRM record, another to book a slot, and Vapi handles the audio stream around it. You can swap the LLM or voice provider later without rewriting the call logic.

    Full entry
  2. 2
    P Apps Voice layer

    ElevenLabs

    by ElevenLabs

    Why this: Callers notice voice quality fast. ElevenLabs is the voice layer that passes the spot check, where a natural pause, a slight inflection, or a clean interruption recovery stops the call from feeling like a phone tree. For an inbound qualifier specifically, the conversational AI surface lets you tune the persona without managing a raw TTS pipeline.

    Full entry
  3. 3
    O OSS Local orchestration / dev loop

    Pipecat

    by Community

    Why this: If you want to own the full audio pipeline rather than staying inside Vapi's managed layer, Pipecat gives you a Python framework that handles the STT-LLM-TTS composition without managing raw audio buffers. Useful for testing qualification scripts locally before pushing to production telephony, or for custom interruption handling that the managed platforms have not caught up on yet.

    Full entry
  4. 4
    M MCP Caller record store

    Supabase MCP Server

    by Supabase

    Why this: When a call comes in, the agent needs to look up whether this number is a known lead, an existing customer, or unknown. The Supabase MCP server lets the voice agent read and write caller records through a scoped token without shipping a service-role key anywhere near the real-time pipeline. Post-call, the transcript and summary land in the same DB row the CRM reads.

    Full entry
  5. 5
    M MCP Booking

    takumi0706/google-calendar-mcp

    by Various

    Why this: The most common escalation path is booking a callback or a meeting. Wiring Google Calendar through the MCP server means the agent can check real availability and confirm the slot on the call, not just promise to follow up. This removes the single biggest drop-off point in phone qualification flows.

    Full entry
  6. 6
    A Agents Hosted alternative

    Help Genie

    by Enterprise DNA

    Why this: If you want an opinionated, hosted version of this entire stack, Help Genie is the ready-made path. It ships call handling, CRM sync, and a Genie training interface with no infrastructure work. Worth knowing when the Vapi-plus-Pipecat custom route is more engineering than the job warrants.

    Full entry
Why we picked this stack

Get this running with Enterprise DNA.

Enterprise DNA closes the loop after the call ends. The transcript and qualification outcome land in the CRM as a structured contact record, the booked escalation shows up in the OPM project for that caller, and the agent's run is logged so you can see which flows are resolving calls versus escalating everything. Secrets for the Vapi API key, ElevenLabs token, and Supabase connection string all live in Infisical, so there is nothing raw in the codebase. The operator inbox surfaces unresolved escalations as a work item, not a voicemail you have to remember to check.

Run this stack on Enterprise DNA
Free Blueprint

Get the Stack Blueprint

A printable architecture card with every tool, role, and rationale on one page.

Enter your email. We send one useful update per week. Unsubscribe any time.

Alternative stacks

Different angles on the same outcome.

Alternative

Outbound qualification agent

If the goal is outbound dialing to pre-qualify a lead list rather than handling inbound calls, the sales outreach stack covers the phone-plus-email motion.

See the alternative Alternative

Email-based support agent

When most of your inbound volume is ticket-based rather than phone-based, the customer support stack covers triage, knowledge retrieval, and CRM updates over email.

See the alternative Alternative

Personal email assistant

If the escalated calls generate emails and you need an assistant that handles the follow-up thread, this stack pairs with the voice agent as a second stage.

See the alternative

Other use cases

More curated stacks from the index.

Use case

Build a customer support agent

A working customer-support agent that triages tickets, answers from your docs, and escalates with full context.

See the stack Use case

Build a research agent

An agent that watches sources, synthesises findings, and ships you a briefing on the days something matters.

See the stack Use case

Build a sales outreach agent

An outreach agent that drafts personal-feeling email, qualifies replies on the phone, and updates the CRM without anyone copy-pasting notes.

See the stack