Docs•Agents Platform

Agents Platform

Build, deploy, and monitor voice agents, live calls, and workflow-driven conversations.

Open featureView API reference
What you get
WAV download • History entry • Reusable asset in your workspace

What it looks like

Agents Platform in SonicVox

Overview

Agents Platform is the control plane for SonicVox voice agents. It combines configuration, workflows, live monitoring, and API control for production conversational systems.

Quickstart

1

Create the agent

Define the name, description, greeting, voice, and behavioral boundaries.

2

Connect workflows and knowledge

Attach the relevant conversation logic, documents, or integrations before deployment.

3

Deploy and monitor

Start the agent, watch live calls, and refine the experience using call history and analytics.

Who this is for

Ideal users

  • Product and operations teams deploying voice agents into real workflows
  • Founders and solution engineers testing call flows before production rollout
  • Teams combining voice, knowledge, and workflow logic in one control plane

Before you start

  • A stable voice choice and a clear agent greeting
  • A documented escalation path for failures or human handoff
  • Knowledge assets, prompts, or workflows prepared before live traffic begins

Settings

SettingDescriptionValues
GreetingThe first line a caller hears when the agent answers.—
Response lengthBalances speed against detail in generated replies.—
Escalation triggersDefines conditions for human handoff or special routing.—

Use cases

Inbound support triage

Answer common questions, collect structured information, and escalate only the calls that truly need a human.

Outbound or scheduled conversational workflows

Use controlled prompts, knowledge, and tools to deliver repeatable conversations with measurable outcomes.

Internal agent prototyping

Test voice identity, workflow branching, and analytics before exposing the experience to end users.

Best practices

  • Keep system prompts concise and focused on actual business boundaries.
  • Review live and historical calls regularly before widening traffic.
  • Treat workflows and prompts as versioned production assets.

Troubleshooting

The agent sounds capable but still fails real calls

Likely cause

Prompts often look good in isolation while escalation logic, knowledge quality, and edge cases remain under-specified.

What to do

Audit the full system: greeting, tool access, workflow branches, escalation triggers, and knowledge freshness before concluding that the model is the only problem.

Live monitoring shows quality drift after deployment

Likely cause

Successful agent behavior depends on stable prompts, curated workflows, and regular review of real conversations.

What to do

Treat agent configuration as versioned production infrastructure. Review calls, tighten prompts, and update workflows or knowledge sources on a recurring cadence.

FAQs

What should I finalize before enabling live traffic?

Finalize the greeting, the escalation policy, the default voice, the core workflow path, and the knowledge base or tool surface the agent is allowed to use.

How do I improve an agent after launch?

Use live monitoring and call history as the source of truth, then refine prompts, workflows, and knowledge instead of guessing from isolated tests only.

How fast is the agent?

Low-latency, real-time turn-taking. Customers experience natural conversation, not robotic pauses.

Can it call my CRM, calendar, knowledge base?

Yes. Built-in integrations for common systems plus a generic webhook/HTTP tool for anything else. Tool calls happen mid-conversation in real time.

What voice does the agent use?

Any voice in your library — a designed brand voice, a cloned spokesperson, or one of 100+ premium voices.

Can I test before going live?

Yes. The test harness runs scripted conversations against your agent, scores performance, and surfaces regressions before you ship.

How do I build the agent flow?

Drag-and-drop workflow editor with branching, conditionals, escalations, and human handoff. Or start from a template (support, sales, scheduling, qualification).

Does it handle handoff to humans?

Yes. Configurable triggers (intent, sentiment, escalation keywords) hand off to a queue with full conversation context.

What about compliance?

Audit trails on every turn, role-based access, PII redaction, and on-prem deployment options. Contact us to discuss the specific regulatory requirements for your deployment.

How does it scale?

Concurrent capacity auto-scales. You pay for the minutes you use — handle a thousand calls or a million without capacity planning.

Detailed guide

Long-form notes, richer formatting, and implementation context for teams that need more than the quickstart.

Deep dive
Rich formatted reference
Use this section for implementation nuance, workflow depth, and operational guidance that does not fit in a simple checklist.

AI Agents Platform Guide

Build voice-enabled AI assistants that can handle phone calls and conversations.

Invite only. Agents is closed to new users while the platform is rebuilt. Unless your workspace has been let in, opening Agents redirects to the waitlist page at /app/agents-waitlist, which explains the rebuild and sends you back to the dashboard. There is no self-serve signup form — access is granted per workspace, so contact support from /app/support if you want to be on the list. Everything below describes the product as it behaves once you have access.

What are AI Agents?

AI Agents are voice-enabled assistants that can:

  • Answer phone calls
  • Have natural conversations
  • Access knowledge bases (RAG)
  • Follow custom workflows
  • Integrate with external services

Creating an Agent

Step 1: Create New Agent

Navigate to Agents → Create Agent

Step 2: Configure Basics

  • Name - Agent identifier
  • Description - What the agent does
  • Greeting - First message to callers

Step 3: Select Voice

Choose from available voices or use a cloned voice.

Step 4: Set Personality

Write a system prompt defining:

  • Agent's persona
  • How to respond
  • Knowledge boundaries

Knowledge Base (RAG)

Add documents to teach your agent:

  1. Go to Knowledge tab
  2. Upload PDF, DOCX, or TXT files
  3. Wait for indexing
  4. Agent will reference these during conversations

Deployment

Start Agent

Click Deploy on the agent page to make it live. The same button pauses the agent again once it is active.

Connect Phone Number

Assign a phone number to receive calls.

Test

Use the Test Call feature to verify behavior.

Features

FeatureDescription
Real-time transcriptionLive conversation text
Call recordingAudio logs for review
AnalyticsCall duration, success rate
WebhooksIntegration events

Configuration Options

  • Interruption Sensitivity - How easily users can interrupt
  • Response Length - Concise, balanced, or detailed
  • Escalation Triggers - Keywords for human handoff
  • Prohibited Topics - Topics to avoid

Best Practices

  1. Keep greetings short - Users want quick responses
  2. Test different prompts - Iterate on personality
  3. Add FAQs to knowledge base - Common questions answered instantly
  4. Monitor call logs - Improve based on real conversations
Was this page helpful?
Agents Platform | SonicVox Docs | SonicVox Docs