Skip to main content

​
What is Response Eagerness?

When a user finishes speaking, the agent needs to decide when to respond. This requires determining whether the user has actually finished their turn or is simply pausing mid-thought. Getting this timing wrong has consequences:
  • Too early — The agent interrupts the user, cutting off their thought and creating a frustrating experience
  • Too late — The conversation feels sluggish and unnatural, with awkward silences after the user finishes speaking
Response Eagerness adjusts this timing to match the conversational context, helping the agent respond at the right moment. Note that response timing also interacts with Thinking Effort. If thinking effort is set to deep, the additional reasoning time may extend the overall response time regardless of eagerness setting.

​
Eagerness Levels

LevelDelay AddedBest For
keenMinimalQuick confirmations, yes/no questions
normalModerateStandard conversation, most steps
patientExtendedComplex questions, thoughtful responses

​
How It Works

User stops speaking
   Voice activity detection confirms silence
   Eagerness delay timer starts
      User speaks again → Timer resets
      Timer expires → Allow Assistant to speak
The system also analyzes whether the user’s thought is complete. Even with “keen” eagerness, if the AI detects an incomplete sentence, it waits longer.

​
When to Use Each Level

​
Keen

Use for:
  • Yes/no questions
  • Simple confirmations
  • Quick acknowledgments
  • Follow-up questions in rapid exchanges
Agent: "Would you like me to send you a confirmation email?"
User: "Yes please"

​
Normal (Default)

Use for:
  • Most conversational turns
  • Standard questions
  • General discussion
Agent: "What brings you to us today?"
User: "I've been looking for..." [moderate response expected]

​
Patient

Use for:
  • Open-ended questions
  • Complex topics
  • When users need to think
  • Important decisions
Agent: "What's been your biggest challenge with your current solution?"
User: "Well, there are a few things..." [extended response expected]

​
Turn Completion Detection

Voxworks doesn’t just use timers. It also analyzes whether the user has completed their thought:
User SaysAnalysisAction
”Yes, that works for me.”Complete sentenceRespond
”I think that…”Incomplete thoughtWait more
”Tuesday is good, but”Trailing conjunctionWait more
”Hmm, let me think.”Explicit pauseWait more
This prevents interrupting users mid-thought, even with faster eagerness settings.

​
First Turns Special Handling

The first two conversational turns use shorter delays:
  • Users expect quick acknowledgment at call start
  • Builds confidence that the system is responsive
  • Prevents awkward silence after greetings
After the initial turns, standard eagerness settings apply.

​
Per-Step Configuration

Set eagerness per step to match the expected interaction: 
Step: "Would you like to proceed?"
  Eagerness: keen
  (Yes/no answer expected)

Step: "Tell me about your current workflow."
  Eagerness: patient
  (Detailed response expected)

Step: "What time works best for you?"
  Eagerness: normal
  (Short but thoughtful answer)

​
Balancing Speed and Accuracy

SettingSpeedRisk
KeenQuick responsesMay interrupt complex thoughts
NormalBalancedGood default for most cases
PatientPatientMay feel sluggish for simple questions
Start with normal and adjust based on testing.

​
Interaction with Other Settings

Eagerness works alongside other dynamics:
Combined WithEffect
Deep thinking effortWait for response + deeper reasoning
Patient silence tolerancePatient waiting before prompting

​
Best Practices

  1. Start with normal — Begin with normal eagerness as your default. Use keen for steps where you want faster responses, or patient for steps where the assistant is jumping in too early or where users might hesitate while thinking
  2. Configure per step — Set eagerness on each step based on expected response complexity
  3. Match eagerness to question complexity — Complex questions need patient eagerness
  4. Test with real conversations — Timing feels different in practice
  5. Consider your users — Some audiences prefer more deliberate pacing

​
Next Steps