Getting started

Build voice agents that pick up the phone

Vollo is an AI voice platform: design conversational agents in a visual workflow builder, put them on real phone numbers or in the browser, and drive everything from a REST API.

What Vollo is#

A Vollo workspace contains projects. Inside a project you create agents (a system prompt plus model, voice, and transcription settings) and workflows — node graphs built in the visual builder that decide what happens on a call: greet, listen, reason with an LLM, look things up, call your APIs, collect keypad digits, transfer to a human, hang up.

A published workflow can be reached three ways:

• Inbound calls — attach a phone number (via a SIP trunk) and route calls to it with dispatch rules.
• Outbound calls — POST /calls starts a call from one of your numbers.
• Web sessions — mint a session token and connect a browser microphone over WebRTC.

Architecture#

Media and control are separate planes. Audio never touches the REST API: it flows through the LiveKit-based media plane straight to the agent workers that execute your workflow and stream audio to the speech and language providers.

The Vollo platform: media plane (left to right) and control plane (bottom).

Quickstart#

From zero to a live voice agent in seven steps. Steps 1–5 happen in the dashboard; the rest is API.

1. Create an account and an API key

   Sign up at app.vollo.io, then create a workspace API key under Settings → API Keys. Keys look like pk_live_… and carry scopes. Store the key server-side — it is shown once.

2. Create a project

   Projects group agents, workflows, numbers, and sessions — one per product or environment is typical. Create one in the dashboard or via POST /projects.

3. Create an agent

   An agent bundles a system prompt with model choices: LLM provider and model, Deepgram for speech-to-text, a Cartesia voice for text-to-speech. You can also configure these inline on a workflow's Agent node.

4. Build a workflow

   Open the visual builder and drag nodes onto the canvas: a Greeting, an Agent node that references your agent, and a Hangup. Wire them start → greeting → agent → hangup. See Workflows for the concepts.

5. Validate and publish

   Press Validate, fix anything it flags, then Publish. Publishing freezes the draft into an immutable version that live traffic runs on — you keep editing the draft safely.

6. Attach a phone number — or open a web session

   For phone: connect a SIP trunk, add a number, and point a dispatch rule at your workflow. For the browser: mint a session token and connect with the LiveKit JS SDK.

7. Make your first API call

   Start an outbound call from your number to any phone:

   curl -X POST https://api.vollo.io/api/v1/calls \
     -H "Authorization: Bearer pk_live_8f3KJd92mA4qL7Zx1RttVWpc" \
     -H "Content-Type: application/json" \
     -d '{
       "project_id": "9b2e7c1a-4d3f-4e8a-9c5b-2f7d1e6a8b3c",
       "workflow_id": "7f3a9d2e-8c1b-4f6a-b5d4-e2c9a1f7b8d0",
       "from": "+12125550142",
       "to": "+962790123456",
       "metadata": { "crm_id": "A-1042" }
     }'

   {
     "data": {
       "id": "c4d9e2f7-1b8a-4e3c-9f6d-7a2b5c8e1d4a",
       "direction": "outbound",
       "status": "queued",
       "from": "+12125550142",
       "to": "+962790123456",
       "project_id": "9b2e7c1a-4d3f-4e8a-9c5b-2f7d1e6a8b3c",
       "workflow_id": "7f3a9d2e-8c1b-4f6a-b5d4-e2c9a1f7b8d0",
       "workflow_version": 3,
       "metadata": { "crm_id": "A-1042" },
       "created_at": "2026-08-18T09:12:44Z"
     }
   }

   The callee's phone rings; when they answer, your published workflow takes over. Track progress with webhooks (call.started, call.ended) or by polling GET /calls/{id}.

Where to next#

• Authentication — scopes, key rotation, and why keys never belong in a browser.
• Workflows — drafts, versions, variables, expressions, and templates.
• Web voice sessions — voice agents in your web app, no phone number needed.
• Webhooks — event delivery, signatures, and retries.
• Errors — the error envelope, rate limits, and quota responses.