> ## Documentation Index > Fetch the complete documentation index at: https://docs.burki.dev/llms.txt > Use this file to discover all available pages before exploring further. # Call Management > Configure interruption handling, timeouts, and call flow control for natural conversations **Think of this as traffic control for your phone calls!** These settings determine how smoothly conversations flow between your AI and callers. *** ## What is Call Management? Call Management controls the **conversation flow** during phone calls. It's separate from AI Configuration (which handles what the AI says) - this section controls **when** and **how** the AI responds to callers. Just like traffic lights control car flow, these settings control conversation flow between your AI assistant and callers. *** ## Interruption Settings **Most Important Section**: These settings directly affect how natural your conversations feel! ### Interruption Style The dashboard exposes three presets: | Style | Best for | Behavior | | ---------------- | -------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------- | | Fast barge-in | Sales, demos, fast support | Stops playback quickly when the caller starts speaking. With Deepgram Flux, playback stops from the provider's speech-start signal. | | Balanced | Most assistants | Filters filler sounds and short noise before yielding. | | Patient listener | Healthcare, senior callers, noisy rooms, IVR replacement | Waits longer before treating caller speech as a real interruption. | The advanced fields below override the selected preset for providers that do not have native turn detection. ### Interruption Threshold **Simple Explanation**: How many words a caller needs to say before the AI stops talking and listens. **Technical Details**: * **Measured in**: Number of words * **Default**: 3 words * **Range**: 1-10 words (recommended) **Real Example**: * Threshold = 3: Caller says "Wait, I have a..." β†’ AI stops talking * Threshold = 1: Caller says "Um..." β†’ AI stops talking (very sensitive) * Threshold = 5: Caller says "Actually, I need to tell you..." β†’ AI stops talking **When to Adjust**: * **Lower (1-2 words)**: For very responsive, human-like conversations * **Higher (4-5 words)**: For formal business calls where AI should finish sentences * **Much higher (6+ words)**: For presentations or when AI shouldn't be interrupted **Technical Note**: The system counts words by splitting text on spaces. Both word count AND min speaking time must be met for a valid interruption. ### Min Speaking Time **Simple Explanation**: How long the AI must talk before it can be interrupted. **Technical Details**: * **Measured in**: Seconds (decimals allowed like 0.5) * **Default**: 0.5 seconds * **Range**: 0.1-3.0 seconds (recommended) **Real Example**: * Min Speaking Time = 0.5s: AI says "Hello, how can Iβ€”" β†’ Caller can interrupt * Min Speaking Time = 2.0s: AI says "Hello, how can I help you today? I'm here to assist withβ€”" β†’ Now caller can interrupt **When to Adjust**: * **Lower (0.1-0.3s)**: For very natural, back-and-forth conversations * **Higher (1-2s)**: For business calls where AI should complete thoughts * **Much higher (2-3s)**: For announcements or important information delivery **Technical Note**: The AI must be speaking for at least this duration before any interruption is considered valid, regardless of word count. ### Interruption Cooldown **Simple Explanation**: How long to pause after a caller interrupts before the AI starts talking again. **Technical Details**: * **Measured in**: Seconds (decimals allowed) * **Default**: 2.0 seconds * **Range**: 0.5-5.0 seconds (recommended) **Real Example**: * Cooldown = 2.0s: Caller interrupts β†’ AI pauses for 2 seconds β†’ AI responds with buffered speech * Cooldown = 0.5s: Very quick responses (might feel rushed) * Cooldown = 4.0s: Longer pauses (might feel awkward) **Technical Note**: During cooldown, any additional speech from the caller is buffered and processed together when the cooldown period ends. **When to Adjust**: * **Lower (0.5-1s)**: For fast-paced conversations or tech support * **Higher (3-4s)**: For emotional or sensitive conversations * **Much higher (5s+)**: For elderly callers or those needing processing time ### Active Listening and Incomplete Utterances **Active listening** enables short backchannel acknowledgments such as "uh-huh" or "mm-hmm" while the caller is speaking. Use the frequency control to keep these cues rare, occasional, moderate, frequent, or aggressive. **Incomplete utterance detection** helps prevent the assistant from answering too early when a caller says a partial phrase like "I want to..." or "Can you help me with...". **Conversation continuity** lets calls from the same number within a short window continue with recent context instead of starting completely fresh. **Natural Conversations**: Threshold=2, Speaking=0.3s, Cooldown=1.5s **Business Formal**: Threshold=4, Speaking=1.0s, Cooldown=2.5s **Customer Support**: Threshold=3, Speaking=0.5s, Cooldown=2.0s (default) *** ## Timeout Settings ### Idle Timeout **Simple Explanation**: How long to wait for the caller to respond before asking "Are you still there?" **Technical Details**: * **Measured in**: Seconds * **Default**: 30 seconds * **Range**: 10-120 seconds (recommended) **Real Example**: * Timeout = 30s: AI asks question β†’ 30 seconds of silence β†’ AI says idle message * Timeout = 10s: Very short patience (good for quick interactions) * Timeout = 60s: Very patient (good for complex questions) **When to Adjust**: * **Lower (10-20s)**: For simple, quick interactions * **Higher (45-60s)**: For complex questions or elderly callers * **Much higher (90s+)**: For forms or detailed information gathering **Technical Note**: The timer resets when the caller completes an utterance (speech\_final) or when the AI responds with meaningful content. ### Max Idle Messages **Simple Explanation**: How many "Are you still there?" messages to send before ending the call. **Technical Details**: * **Measured in**: Number of messages * **Default**: 3 messages * **Range**: 1-5 messages (recommended) **Real Example**: * Max = 3: AI asks "Are you there?" β†’ Silence β†’ AI asks again β†’ Silence β†’ AI asks again β†’ Silence β†’ Call ends * Max = 1: AI asks once, then ends call (not very patient) * Max = 5: AI is very patient (might annoy callers) **When to Adjust**: * **Lower (1-2)**: For business efficiency or when call time is expensive * **Higher (4-5)**: For customer service or when you don't want to lose calls **Quick Service**: Idle=15s, Max Messages=2 **Standard Support**: Idle=30s, Max Messages=3 (default) **Patient Service**: Idle=45s, Max Messages=4 *** ## Call Control Messages ### End Call Message **Simple Explanation**: The last thing callers hear when the call ends (either by AI decision or timeout). **Default**: "Thank you for calling. Goodbye!" **Examples**: * Business: "Thank you for calling \[Company]. Have a great day!" * Support: "Your issue has been resolved. Thank you for calling!" * Appointment: "Your appointment is confirmed. We'll see you soon!" **Best Practices**: * Keep it under 10 words for clarity * Include your company name if branding is important * Match the tone of your assistant's personality ### Transfer Call Message **Simple Explanation**: What the AI says when transferring a caller to a human agent. **Default**: "Please hold while I transfer your call." **Examples**: * Professional: "Let me connect you with a specialist who can help." * Urgent: "I'm transferring you to our emergency support team now." * Specific: "Connecting you to our billing department. Please hold." **Best Practices**: * Explain WHO they're being transferred to * Keep them calm with reassuring language * Include estimated wait time if known ### Idle Message **Simple Explanation**: What the AI says when there's been silence for too long. **Default**: "Are you still there? I'm here to help if you need anything." **Examples**: * Friendly: "Hello? I'm still here if you have more questions!" * Professional: "Are you still on the line? I'm ready to assist." * Helpful: "Take your time. Let me know when you're ready to continue." **Best Practices**: * Sound patient, not annoyed * Offer help or reassurance * Keep it conversational and warm *** ## Recording Settings **Configurable per assistant**: Recording can be enabled or disabled from the Safety section. When enabled, Burki can store mixed call audio, optionally store a caller-only track, and play a disclosure message before recording. Recording-related settings now live in **Safety**: * **Enable recording:** store call audio for review and compliance. * **User-only track:** use organization default, force on, or force off for a caller-only recording. * **Recording disclosure:** play a disclosure message every call or only the first call per caller. *** ## Common Configuration Scenarios ### Customer Service **Goal**: Helpful but efficient support ```yaml theme={null} Interruption Threshold: 3 words Min Speaking Time: 0.5 seconds Interruption Cooldown: 2.0 seconds Idle Timeout: 30 seconds Max Idle Messages: 3 ``` **Why**: Balanced responsiveness with professional patience ### Sales Calls **Goal**: Let prospects talk, show you're listening ```yaml theme={null} Interruption Threshold: 2 words Min Speaking Time: 0.3 seconds Interruption Cooldown: 1.5 seconds Idle Timeout: 45 seconds Max Idle Messages: 4 ``` **Why**: Very responsive to show engagement, patient to let them think ### Appointment Booking **Goal**: Clear information exchange ```yaml theme={null} Interruption Threshold: 4 words Min Speaking Time: 1.0 seconds Interruption Cooldown: 2.5 seconds Idle Timeout: 20 seconds Max Idle Messages: 2 ``` **Why**: AI needs to finish instructions, but move efficiently through booking ### Tech Support **Goal**: Handle frustrated callers patiently ```yaml theme={null} Interruption Threshold: 2 words Min Speaking Time: 0.4 seconds Interruption Cooldown: 3.0 seconds Idle Timeout: 40 seconds Max Idle Messages: 4 ``` **Why**: Very responsive to interruptions, longer cooldowns for emotional regulation *** ## Troubleshooting Common Issues **Problem**: Interruption threshold too low or min speaking time too short **Solutions**: * Increase interruption threshold to 4-5 words * Increase min speaking time to 1.0-1.5 seconds * Callers won't interrupt as quickly **Test**: Make a call and see if AI can complete full sentences **Problem**: Interruption cooldown too long or threshold too high **Solutions**: * Decrease interruption cooldown to 1.0-1.5 seconds * Decrease interruption threshold to 2-3 words * AI will feel more responsive and natural **Test**: Try interrupting mid-sentence - AI should stop and respond quickly **Problem**: Idle timeout too short or max idle messages too low **Solutions**: * Increase idle timeout to 45-60 seconds * Increase max idle messages to 4-5 * Give callers more time to think and respond **Test**: Stay silent during a call - should get multiple "are you there?" messages **Problem**: Idle timeout too long or max idle messages too high **Solutions**: * Decrease idle timeout to 15-20 seconds * Decrease max idle messages to 1-2 * Calls will end more efficiently **Test**: Simulate an unresponsive caller - call should end reasonably quickly *** ## Best Practices 1. **Start with defaults** and adjust based on real call feedback 2. **Test with real users** - your perception differs from callers' 3. **Match your business style** - formal vs casual affects all settings 4. **Monitor call recordings** to see where conversations break down 5. **Adjust gradually** - small changes have big impacts on conversation feel **Pro Tip**: Record yourself having a conversation with your assistant, then listen back. You'll immediately hear what needs adjusting! *** **Ready to fine-tune your AI's brain?** Check out [AI Configuration](/ai-configuration) to optimize your LLM, TTS, and STT settings. **Need better speech detection?** See [STT Providers & Settings](/stt-providers) for endpointing, VAD, and silence threshold tuning. **Need live monitoring?** See [Live Transcript WebSocket](/live-transcript) for real-time call monitoring and transcript streaming. **Need voice- and audio-specific tuning?** Use [Voice tuning](/tts-providers/voice-tuning) for TTS controls, then refine STT/TTS and interruptions in [AI Configuration](/ai-configuration).