Voice Agents
Voice agents are the core of HubTalk AI. Each agent combines a language model (LLM), speech recognition (ASR), and text-to-speech (TTS) to hold real-time conversations with callers.
Agent List
The Agents section displays all agents in your organization. You can search by agent name or company name. Each agent shows its current configuration and status.
Creating an Agent
To create a new agent, click Create Agent in the top right corner. The agent editor has two main areas:
Component Selection
At the top of the agent editor, select the three core components:
| Component | Purpose |
|---|
| LLM | The language model that powers the agent’s reasoning and responses |
| ASR | Automatic Speech Recognition — converts caller’s speech to text |
| TTS | Text-to-Speech — converts the agent’s text responses to spoken audio |
System Prompt
Below the component selector is the main prompt editor. HubTalk AI uses a single prompt model — you write one comprehensive system prompt that defines the agent’s personality, instructions, constraints, and behavior.
Write your prompt in the language you want the agent to use by default. The agent can switch languages during conversation if instructed.
Greeting Configuration
Configure how the agent greets callers:
- Fixed greeting — A static message spoken at the start of every call
- Dynamic greeting — A greeting that incorporates variables (caller name, time of day, campaign data)
The greeting is synthesized via TTS and played immediately when the call connects, before the LLM processes any input. This reduces perceived latency.
Right Panel Settings
The right panel provides access to advanced configuration:
| Setting | Description |
|---|
| Functions | Add operator transfer, call ending, and custom API functions |
| Post-call Analysis | Configure variables extracted after each call |
| VAD Settings | Fine-tune voice activity detection (interruption, silence, speech thresholds) |
| Speech Settings | Configure agent behavior during caller silence |
| Webhook | Set the URL where call data is sent after each conversation |
| Knowledge Base | Attach document collections the agent uses to answer questions |
VAD Settings
Voice Activity Detection (VAD) controls how the agent handles interruptions, pauses, and speech boundaries.
| Parameter | Description |
|---|
| Interruption threshold (first/last slider) | Controls how the agent reacts when the caller starts speaking over the agent |
| Max speech pause | How long the agent waits silently if the caller pauses mid-dialog |
| Min speech counter | Minimum speech length for the system to count it as “real” speech |
| Interruption min speech counter | How much “confident speech” is required before the caller can interrupt the agent |
| Silence counter for buffer reset | After prolonged silence, the system clears the accumulated audio buffer |
All parameters are measured in audio chunks, except for the interruption settings.
Speech Settings
Speech settings control how the agent responds to caller silence. For example, you can configure the agent to ask “Are you still there?” every 7 seconds of silence, up to 3 times.
| Parameter | Description |
|---|
| Silence interval | Seconds of silence before the agent speaks |
| Max prompts | Number of times the agent prompts the caller before taking action |
Chat Testing
Every agent has a built-in chat testing feature accessible from the agent editor. Use it to:
- Test prompt behavior with simulated conversations
- Verify function triggers and responses
- Check knowledge base retrieval
- Validate multilingual switching
Chat testing simulates the text layer only. For full voice testing (ASR + TTS), make a real test call through the telephony system.
