My Agent Doesn't Speak or Goes Silent
Find out why an agent says nothing when a call connects, or stops talking partway through, and fix it in the right setting.
Written By voiceaiwrapper
Last updated 26 days ago
When a caller hears nothing, the cause is usually a setting, not an outage. This article walks you through the checks in order, from the most common cause to the least, so you can find the problem in a few minutes and fix it in the right place.
Start with the call itself. The call's record tells you which of the checks below applies.
Step 1: Open the call in Call Logs
- In the sidebar, open Call Logs for the Pod (an engine or provider connection) the agent belongs to.
- Find the call and click it.
What to look at depends on the engine:
- VAW Native: if the call failed, the Overview tab shows a red box at the top with the end reason and an error code. The End reason row under Metadata says how the call finished. The What ran tab shows the model, voice, language and time limit the call actually used, taken when the call started.
- Vapi, Retell AI and ElevenLabs: check the End Reason column and the transcript. The end reason comes from your provider.
If you can't find the call at all, see Calls Are Missing from Call Logs.
Use this table to jump to the right check:
Check 1: Credits (VAW Native) or provider balance (BYOK)
VAW Native. Calls on VAW Native use AI credits. When your balance runs out, new calls can't start. A call that's already running is never cut off, but the next one is refused:
- Inbound callers hear a busy tone.
- Test calls and web calls don't start.
- Refused inbound calls and test calls appear in Call Logs with the end reason Insufficient credits.
You'll also see a red banner at the top of the webapp, You are out of credits (balance: β¦), with "New calls are paused until you top up." To fix it, click Buy credits in the banner, or open Settings β Billing and use Buy credits on the Credits card. Buying credits charges your card. See What Happens When You Run Out of Credits.
Vapi, Retell AI and ElevenLabs (BYOK Pods). Calls run on your own provider account and use its balance, not VoiceAIWrapper credits. If that account is out of funds or its billing failed, the provider can refuse or end calls. Sign in to your provider's dashboard and check its billing page.
Check 2: Who speaks first and the greeting
This is the most common cause of a call that connects in silence.
VAW Native. Open the agent. On the Agent Settings tab, look at How do you want your agent to greet callers? and the Who speaks first switch next to it:
- Caller speaks first: the agent waits for the caller to talk before it says anything. On inbound calls and web calls, the caller usually waits too, so nobody speaks. Switch to Agent speaks first for agents that answer calls.
- Agent speaks first with an empty greeting: the agent has no line to say, so it waits for the caller. Type a greeting.
An empty greeting field shows "Hello! How can I help you today?" in grey. That's a placeholder, not a greeting. If the text is grey, the field is empty.
Click Save changes after you edit. Calls always use the last saved version of the agent. See Write the Prompt and Greeting.
Retell AI. In Agents, click Edit prompt on the agent and check Welcome Message Type. User Initiates: AI remains silent until users speak first. means the agent waits. Choose one of the AI Initiates options for agents that answer calls, then click Update Agent.
Vapi and ElevenLabs. In the Edit prompt sheet, check that First Message says what you expect. Other first-message options, such as Vapi's setting for the assistant to wait for the user, are only in your provider's dashboard. See Edit Prompts and Greetings for Provider Agents.
Check 3: The voice (VAW Native)
Some older agents still use one of the retired OpenAI voices: Alloy, Ash, Ballad, Coral, Echo, Fable, Nova, Onyx, Sage, Shimmer or Verse. These voices sometimes produce no audio, so the agent goes silent or answers very late, with no error on the call. They can no longer be picked, but agents that already had one keep it.
To fix it:
- Open the agent and, on Agent Settings, click the Voice row.
- Pick a voice from the list. Every voice you can pick there is an ElevenLabs voice.
- Close the sheet and click Save changes.
See Choose a Voice for Your Agent.
Check 4: The language (VAW Native)
The Language setting tells the agent which language callers will speak. If it's set to one language and the caller speaks another, the agent may not understand what the caller says. It can then seem to go silent after the greeting.
- On Agent Settings, click the Language row.
- Pick the language your callers speak, or Auto-detect. Auto-detect covers English, Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian and Dutch. For any other language, pick it from the list.
- Click Save changes.
If the voice isn't made for the language you chose, the Voice sheet warns that callers may hear a slight accent. That affects how the agent sounds, not whether it speaks. See Choose the Agent's Language.
Check 5: Response patience and interruptions (VAW Native)
On Agent Settings, click the Conversation row:
- Response patience sets how long the agent waits after the caller stops talking before it replies. Patient waits longest, which can feel like a pause on every turn. Try Balanced or Fast.
- Allow interruptions and Interruption sensitivity let callers cut the agent off. On a noisy line with High sensitivity, background noise can keep stopping the agent mid-sentence. Lower the sensitivity, or turn on Background noise filter.
Save your changes and test again. See Fine-Tune Conversation Behaviour.
Check 6: Model, voice or transcriber errors (VAW Native)
If the end reason is LLM failed, Voice failed or Transcriber failed, a service the call depends on returned an error during the call. The red box on the call's Overview tab shows the error code and message.
- Open the What ran tab to confirm which Model and Voice the call used. If it isn't what you expect, you may have edited the agent without saving.
- If one call failed, place a test call. A single failure is often temporary.
- If every call fails with LLM failed, try a different model in the AI model sheet, save and test again. See Choose an AI Model.
- If Voice failed repeats, switch to a different voice (Check 3).
If the error keeps happening on every model or voice, contact support (see the end of this article).
Check 7: The silence timeout ended the call (VAW Native)
End call after silence on the agent's Call Settings tab hangs up when the caller has been quiet for the time you choose. The agent says a short goodbye first. The call ends with Silence timed out.
If calls end while callers are still thinking, or reading out details, choose a longer time or Off, then save. The timer starts when the caller is first heard, so ringing time doesn't count. See Call Settings: Recording, Silence Timeout and Maximum Call Length.
Check 8: A slow tool (VAW Native)
When the agent runs a tool, such as a calendar lookup, it waits for the answer. If the tool is slow, the caller hears silence.
- Give the agent something to say while it waits. On the agent's Tools tab, open the attached tool and fill in What β¦ says while it runs (the label includes your agent's name), for example "Let me check that for you." Click Save changes in the sheet. It applies from the next call.
- Check how long the agent waits. In the sidebar, open Tools, click the tool and open its Safety tab. Timeout is the longest the agent waits for an answer: 1 to 30 seconds, 10 by default. Lower it if your system usually answers quickly.
See Tool Safety, Caller Confirmation and Timeouts.
Check 9: Webhook sync (Vapi, Retell AI and ElevenLabs)
An unsynced agent doesn't send its calls to VoiceAIWrapper, so you can't see the transcript or the end reason here. Open Assistants (Vapi) or Agents (Retell AI, ElevenLabs). If the agent's row shows Sync required, click Sync now, then place another test call so you have a record to look at. See Sync Webhooks So VoiceAIWrapper Can Track Calls.
Check 10: Errors inside your provider (Vapi, Retell AI and ElevenLabs)
The voice, model and transcriber of a provider agent are set in your provider's dashboard, and errors there aren't shown in full in VoiceAIWrapper. Copy the provider's call ID from the call log row, then look the call up in your provider's call logs. Common causes are:
- A model, voice or custom LLM that returned an error or timed out.
- A voice that doesn't support the language the agent speaks.
- A custom function or tool server that didn't answer.
Fix the setting in your provider's dashboard, then test again.
When to contact support
If you've worked through the checks and the agent is still silent, open Need help? β Help Chat and include:
- The Call id from the call's Overview tab (VAW Native) or the provider's call ID (Vapi, Retell AI, ElevenLabs).
- Your Tenant ID from Settings β General ("Reference this when contacting support.").
- The agent's name, and roughly when the call happened.
- What you've already checked.
See Getting Help: Support Chat, Book a Call and Feature Requests.
Related articles
Was this helpful?
More in Troubleshooting and FAQs
My Outbound Campaign Isn't CallingCalls Are Missing from Call LogsMy Custom Domain Won't VerifyCSV Import Errors and Invalid LeadsStill need help? Share an idea