One agent contract. Browser and phone conversations.
Keep your business logic independent of the voice runtime. VoiceForge routes supported sessions through an adapter interface, with access determined by your plan.
Describe & review
Make the instructions your own
Test a conversation
Hear it before you launch
Publish a version
Improve from real outcomes
Caller audio
Browser or configured phone connection.
Understand
Speech recognition for your conversation.
Respond
Reasoning with configured knowledge and tools.
Speak
Voice synthesis and return audio.
The voice runtime
ElevenLabs Conversational AI handles speech recognition, conversation reasoning, and voice synthesis. VoiceForge keeps the Agent Spec, workspace permissions, and calling controls separate through a provider adapter.
Browser testing
Test a conversation in your browser before connecting a phone number. Runtime admission follows workspace permissions and plan entitlements.
Connect telephony when you are ready
Connect your Twilio account, import the numbers your workspace should manage, and assign a published agent. Verify inbound and outbound paths before launch.
Understand what affects response time
Network conditions, audio turn detection, context size, and tool execution all affect a conversation. Review the diagnostics available for your sessions and test under realistic conditions rather than relying on a single advertised latency figure.
Hear the difference.
Then build your own.
Start with a free workspace. Review your agent. Test before you publish.