VoiceForge AI
Build voice agents from natural language, govern every behavior with Agent Spec JSON, test the call before it goes live, and hand clients a branded workspace they can trust.
5 min
from plain prompt to a tested call
Spec
Agent JSON before every publish
100%
of outbound calls pass compliance gates first
4
production voice provider adapters
From brief to live workflow
Four steps from brief to a monitored live agent.
VoiceForge is not a prompt wrapper. It is a controlled build, test, publish, and monitor loop for teams that need voice agents to survive real customer calls.
Describe the caller experience
Capture the business, caller intents, transfer rules, knowledge sources, and systems the agent may touch.
Review the generated contract
VoiceForge produces Agent Spec JSON with goals, tools, compliance policy, analytics, and handoff behavior.
Test the call before it ships
Run a full call in the browser with live transcript, event stream, and outcome before you point real telephony at it.
Publish, monitor, improve
Ship a demo page, track calls, inspect transcripts, tune versions, and give clients a branded workspace.
Product surface
Show the builder, then prove the guardrails.
One system covers the whole lifecycle: prompt generation, versioned specs, test calls, publish controls, transcripts, analytics, and client branding.
- Natural-language generation with Zod-validated Agent Spec output
- Visual flow editing for questions, transfers, tools, knowledge, and endings
- Agent-scoped knowledge sources for business-specific retrieval
- Versioning, publish controls, and audit logs around critical actions
- Call transcripts, events, outcomes, minutes, and usage analytics
Real-world controls
Built for calls that have legal, brand, and revenue consequences.
Every promise here maps back to an engineering rule: validated specs, scoped data, permissioned tools, adapter boundaries, audit logs, and compliance checks before outbound execution.
Agent Spec JSON
The agent is governed by a reviewable contract, not hidden prompt strings scattered across runtime code.
Compliance gates
Consent, call windows, opt-outs, DNC/DND, AI disclosure, and recording notice run before outbound execution.
Workspace isolation
Organizations, workspaces, clients, agents, calls, knowledge, and settings stay tenant scoped.
Provider adapters
Vapi, Retell, OpenAI Realtime, and Twilio run behind one runtime interface, so switching provider never rewrites the agent.
Permissioned tools
Tool calls are validated, authorized, logged, and idempotent where the action allows it.
Postgres source of truth
Agent versions, workspaces, calls, compliance records, knowledge, and billing usage stay structured.
White-label delivery
Agencies can ship client dashboards with brand colors, logos, isolated users, and focused analytics.
Public demo
Let prospects hear the product before they read the docs.
Play a real 30-second call from a published agent, then read the transcript and the outcome signals your buyers will ask about.
Smile Dental Receptionist
Published demo agent
Dental receptionist - 30 sec call
Generated sample call audio
Thanks for calling Smile Dental. Are you calling to book, reschedule, or ask about an emergency?
I need a cleaning next week, preferably morning.
I can help with that. I found two morning openings and can confirm the best one.
Agency-ready
Sell voice agents without rebuilding the operating layer.
VoiceForge gives agencies and internal operators the pieces that usually delay launch: client isolation, branded delivery, usage visibility, provider choice, and compliance-first operations.
White-label control
Configure logos, colors, workspace settings, and client-facing pages without forking the product.
Client isolation
Client users only see their workspace, calls, agents, knowledge, analytics, and settings.
Integration path
VoiceForge AI