> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tryhamsa.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Call behavior settings

> Tune how your agent behaves on a call: who speaks first, turn taking, interruptions, silence, ending the call, noise, sounds, time zone and caller detection.

Call settings decide how a conversation feels: how long the agent waits before it answers, whether the caller can cut in, what happens on silence and when the call ends. This page explains each setting, what it does on a call, and when to change it.

## Where to find it

> [tab] Single prompt agent
>
> In the editor's **Configuration** panel, open **Call settings**. **Noise cancellation** is its own section in the same panel. **Enhanced turn taking** sits under **Agent instructions**.

> [tab] Flow agent
>
> In the flow builder, click **Global settings** and open **Call settings**. **Noise cancellation** is its own section there. **Enhanced turn taking** is in **System prompt**. A [Change agent settings step](https://docs.tryhamsa.com/agents/flow/steps/change-settings.md) can change call settings from that step on, and each conversation step has its own **Interruption** setting.
>
> ![The flow builder with Global settings open on Call settings, showing Wait for user to speak first, On-hold music, Timezone, Agentic RAG, User gender detection and Smart call end](https://mintcdn.com/hamsa-452fa79a/IyoYQ-nNrFRmlJOT/images/agents/settings/flow-global-call-settings-light.webp?fit=max\&auto=format\&n=IyoYQ-nNrFRmlJOT\&q=85\&s=86e48fb41251cda9cfc7d63fd1362702)

The defaults suit most agents. Change one setting at a time and make a [test call](https://docs.tryhamsa.com/agents/release/test-your-agent.md) after each change.

![Call settings in the single prompt Configuration panel: Wait for user to speak first set to Never, On-hold music, Timezone, User gender detection and Smart call end](https://mintcdn.com/hamsa-452fa79a/IyoYQ-nNrFRmlJOT/images/agents/settings/call-settings-light.webp?fit=max\&auto=format\&n=IyoYQ-nNrFRmlJOT\&q=85\&s=33b150238b31243328055905be61fdae)

## Start of the call

**Wait for user to speak first** decides whether the agent opens the call.

| Option | What happens |
| - | - |
| **Never** (default) | The agent speaks its first line as soon as the call connects. |
| **Always** | The agent stays silent until the caller speaks. The caller's first words start the greeting; the agent doesn't answer them separately. |
| **Outbound calls** | The agent waits only on calls it places to a phone number. On inbound and browser calls it speaks first. |

Use **Outbound calls** when your agent calls people: many people answer with "Hello?" and expect to hear who is calling after that.

## Turn taking and interruptions

These settings control when the agent decides the caller has finished, and whether the caller can talk over the agent.

![Call settings scrolled to Ambient sound, Thinking voice, Speaker identification, Response delay at 400 ms, User inactivity timeout at 15 s and Max call duration](https://mintcdn.com/hamsa-452fa79a/IyoYQ-nNrFRmlJOT/images/agents/settings/call-settings-turn-taking-light.webp?fit=max\&auto=format\&n=IyoYQ-nNrFRmlJOT\&q=85\&s=d87a892d156eeaf320d56dd3895e35c1)

| Setting | What it does | Range | Default |
| - | - | - | - |
| **Interrupt** | Lets the caller interrupt the agent while it speaks. When off, the agent doesn't listen while it speaks. | On or off | On |
| **Response delay** | How long the caller must be silent before the agent treats them as done. Values under 250 ms act as 250 ms. | 100 to 1500 ms, steps of 100 | 400 ms |
| **Minimum interruption duration** | How long the caller must speak over the agent before it counts as an interruption. Shorter sounds don't stop the agent. | 0.2 to 1.5 s, steps of 0.1 | 0.5 s |
| **VAD activation threshold** | How loud and clear sound must be to count as speech. Higher ignores more background noise but can miss quiet speakers. Phone calls use 70% of this value, browser calls 90%. | 0.2 to 0.9, steps of 0.1 | 0.5 |
| **Enhanced turn taking** | Better handling of numbers read out in pieces and of short "mm-hm" sounds while the agent speaks, which are no longer taken as interruptions. | On or off | Off |

When to change them:

* **The agent cuts callers off mid-sentence**, for example while they read a number: turn on **Enhanced turn taking**, or raise **Response delay**.
* **Replies feel slow**: lower **Response delay** in steps of 100 ms.
* **Background noise stops the agent**: raise **Minimum interruption duration** or **VAD activation threshold**, or turn on [noise cancellation](#noise-cancellation).
* **Callers can't stop a long answer**: check **Interrupt** is on.

Some speech is never interruptible: the greeting, the message at the maximum call length, and the goodbye from smart call end. If the caller only made a short sound, the agent picks up where it stopped after half a second.

> [tab] Single prompt agent
>
> **Interrupt** applies to everything the agent says, apart from the lines above.

> [tab] Flow agent
>
> Only conversation steps can be interrupted. Each conversation step has an **Interruption** setting: **Inherit** (use **Interrupt** from Call settings), **Enable** or **Disable**. The Start step's opening message has its own setting. See [Conversation step](https://docs.tryhamsa.com/agents/flow/steps/conversation.md).

## When the caller goes quiet

**User inactivity timeout** (5 to 60 seconds, default 15) is how long the agent waits in silence before it checks in. When the timeout passes, the agent asks whether the caller is still there and if it can help. It asks again after every further timeout. The agent doesn't hang up because of silence, so pair this with a sensible **Max call duration**.

## End the call

### Smart call end

Turn on **Smart call end** to let the agent hang up when the conversation is done. The agent decides from the **Smart call end prompt**, which starts with this text:

> Terminate the call when: (1) all customer inquiries have been resolved, (2) the conversation has reached a natural conclusion, (3) the customer's goal has been achieved, or (4) further conversation would be unproductive. Use this tool only when you've achieved the primary objective of the call.

Edit the prompt to match your calls, for example "End the call after the appointment is confirmed and the caller has no other questions." Click **Restore** to go back to the default text. When the agent ends the call, it says a short goodbye first.

![Smart call end turned on with its Smart call end prompt and the Restore button, followed by Language/dialect switcher, Interrupt, Ambient sound, Thinking voice and Speaker identification](https://mintcdn.com/hamsa-452fa79a/IyoYQ-nNrFRmlJOT/images/agents/settings/call-settings-smart-call-end-light.webp?fit=max\&auto=format\&n=IyoYQ-nNrFRmlJOT\&q=85\&s=7343de6dfe25814b6daa72cf112e1367)

> [tab] Single prompt agent
>
> With **Smart call end** off, a single prompt agent never hangs up on its own; the call ends when the caller hangs up or the maximum call length is reached.

> [tab] Flow agent
>
> A flow also ends at an [End call step](https://docs.tryhamsa.com/agents/flow/steps/end-call.md). **Smart call end** adds a way to end the call from any point of the conversation.

### Maximum call length

**Max call duration** caps every call, from 0.5 to 60 minutes in steps of 0.5. The default is 5 minutes. Raise it for agents that handle long calls.

**When the limit is reached** sets what happens then:

![Max call duration set to 10 minutes, When the limit is reached set to End the call with an optional Message before ending, then Minimum interruption duration and VAD activation threshold](https://mintcdn.com/hamsa-452fa79a/IyoYQ-nNrFRmlJOT/images/agents/settings/call-settings-max-duration-light.webp?fit=max\&auto=format\&n=IyoYQ-nNrFRmlJOT\&q=85\&s=b0732aa4fc4b04a21dfc18453d5105ea)

| Option | What happens |
| - | - |
| **End the call** (default) | The agent says the message, then hangs up. |
| **Transfer to a person** | The agent says the message, then transfers the call to the number you enter. The caller hears on-hold music during the transfer. |

* The message (**Message before ending** or **Message before transferring**) is optional, up to 2,000 characters, and can use variables. If you leave it empty, the agent says a short goodbye explaining that the maximum call time was reached.
* Before ending, the message is said as written. Before a transfer, choose **Static** to say it as written, or **Prompt** to give it to the agent as an instruction, such as "Tell the caller you're passing them to a colleague".
* For a transfer, enter the number in international format (for example `+962791234567`), or a variable such as `{{ manager_number }}`.
* Transfers only work when a phone number is assigned to the agent, so they don't happen on browser calls. If a transfer fails, the call ends.
* The caller can't interrupt the message.

## Noise cancellation

Noise cancellation filters background noise from the caller's audio before the agent listens to it.

> [tab] Single prompt agent
>
> In **Noise cancellation**, pick a **Model**: **Disabled**, **Telephony optimized** or **General use-cases**. The other fields appear when a model is picked.

> [tab] Flow agent
>
> In **Noise cancellation**, turn on **Noise cancellation**, then pick a **Model**: **Telephony optimized** or **General use-cases**. **Enable auto-gain control** is always available.

| Field | Options | Default |
| - | - | - |
| **Model** | **Telephony optimized**, tuned for phone audio, or **General use-cases** | Telephony optimized |
| **Noise cancellation strategy** | **Per conversation**: noise is filtered once for the whole call, best for long continuous conversations. **Per turn**: filtered separately for each turn, best when voices or surroundings change often. | Per conversation |
| **Enable auto-gain control** | Evens out the caller's volume. | Off |
| **Send denoised to STT** | Sends the filtered audio to speech recognition. When off, recognition hears the original audio and only speech detection uses the filtered audio. | Off |

![Noise cancellation in the Configuration panel: Model set to Telephony optimized, Noise cancellation strategy set to Per conversation, and the Enable auto-gain control and Send denoised to STT switches](https://mintcdn.com/hamsa-452fa79a/IyoYQ-nNrFRmlJOT/images/agents/settings/noise-cancellation-light.webp?fit=max\&auto=format\&n=IyoYQ-nNrFRmlJOT\&q=85\&s=a1834d1105f52751a848141a9ab12aea)

Noise cancellation and **Ambient sound** don't affect each other: one cleans what the agent hears, the other adds sound to what the caller hears.

## Sounds the caller hears

| Setting | What the caller hears | Default |
| - | - | - |
| **Ambient sound** | A background sound for the whole call. Pick it in **Ambient sound track**; if none is picked, or the track can't be loaded, an office ambience plays. | Off |
| **Thinking voice** | Keyboard typing sounds while the agent works out its reply or runs a tool. | Off |
| **On-hold music** | The track that plays while the call is on hold: while a tool with on-hold music runs, and during a call transfer. | The platform's default track |

These work on phone and browser calls alike.

### Pick or upload a track

**On-hold music** and **Ambient sound track** each have a track picker. It lists Hamsa's tracks under **Hamsa library** and your project's under **My uploads**. Beside the picker, **Preview** plays the selected track, **Upload music** adds one, and **Delete track** removes the selected upload. Hamsa's own tracks can't be deleted.

![The On-hold music picker open, with Default, Soft piano and Calm strings under Hamsa library and two tracks under My uploads, and Manage library beside it](https://mintcdn.com/hamsa-452fa79a/IyoYQ-nNrFRmlJOT/images/agents/settings/on-hold-music-picker-light.webp?fit=max\&auto=format\&n=IyoYQ-nNrFRmlJOT\&q=85\&s=3f0fe79a60a3d9fda3b4b55816271f8f)

**1. Open the library**

Under the picker, click **Manage library**. The **On-hold music library** (or **Ambient sound library**) lists every track with its length. Hamsa's tracks are marked **Preset**, yours **Uploaded**.

![The On-hold music library: three Preset tracks and two Uploaded tracks with preview, rename and delete buttons, and Upload music at the bottom](https://mintcdn.com/hamsa-452fa79a/IyoYQ-nNrFRmlJOT/images/agents/settings/on-hold-music-library-light.webp?fit=max\&auto=format\&n=IyoYQ-nNrFRmlJOT\&q=85\&s=dbe502ec4c8922616cd6a96ed247e918)

**2. Upload a track**

Click **Upload music**. In **Upload a track**, click **Choose an audio file** and pick an mp3, wav or ogg file of up to 10 MB and 5 minutes, enter a **Track name**, and click **Upload**. Each name can be used once.

![The Upload a track form: Choose an audio file with mp3, wav or ogg, up to 10MB, max 5 minutes, and an empty Track name](https://mintcdn.com/hamsa-452fa79a/IyoYQ-nNrFRmlJOT/images/agents/settings/on-hold-music-upload-light.webp?fit=max\&auto=format\&n=IyoYQ-nNrFRmlJOT\&q=85\&s=c35817264faaab5111dbf6dd21c089dc)

**3. Rename or delete an upload**

In the library, click **Rename** or **Delete track** on one of your uploads.

Uploads belong to the project: "Shared across all agents in this project". A track you upload while picking one is selected for this agent; one you upload in the library is only added to it.

## Know who is speaking

| Setting | What it does | Default |
| - | - | - |
| **User gender detection** | Speech recognition estimates the caller's gender. Each turn, the model is told who it is speaking to and, with the prompt enhancer on, how to use the matching Arabic forms. `{{ user_gender }}` holds the value; it is `unknown` until one is detected. Helpful for Arabic speaking agents. | Off |
| **Speaker identification** (Beta) | Tells apart several people on the same call (Speaker1, Speaker2, and so on). The model is told which speaker is talking each turn and asked to learn their names, without calling them by the label. | Off |

## Time zone

**Timezone** is the time zone the agent works in. It sets the time variables in prompts and messages (`{{ current_time }}`, `{{ current_date }}` and the rest), and the date and time the model is given on every turn.

Set it to your business's time zone: search for a city or region and pick it. If you leave it on **Default (UTC)**, calls use the platform's default time zone, which may not be UTC, so always set it when your agent talks about dates, opening hours or appointments.

## Settings reference

| Setting | Default | Range or options |
| - | - | - |
| **Wait for user to speak first** | **Never** | **Never**, **Always**, **Outbound calls** |
| **On-hold music** | Platform default | Track picker |
| **Timezone** | **Default (UTC)** | Any time zone |
| **User gender detection** | Off | On or off |
| **Smart call end** | Off | On or off, with **Smart call end prompt** |
| **Language/dialect switcher** | Off | On or off. See [Voice, language and dialect](https://docs.tryhamsa.com/agents/settings/voice-and-language.md#let-the-agent-switch-dialect-during-a-call) |
| **Interrupt** | On | On or off |
| **Ambient sound** | Off | On or off, with **Ambient sound track** |
| **Thinking voice** | Off | On or off |
| **Speaker identification** | Off | On or off |
| **Response delay** | 400 ms | 100 to 1500 ms |
| **User inactivity timeout** | 15 s | 5 to 60 s |
| **Max call duration** | 5 min | 0.5 to 60 min |
| **When the limit is reached** | **End the call** | **End the call**, **Transfer to a person** |
| **Minimum interruption duration** | 0.5 s | 0.2 to 1.5 s |
| **VAD activation threshold** | 0.5 | 0.2 to 0.9 |

Flow agents also have **Agentic RAG** in **Call settings**; single prompt agents have it in **Knowledge base**. See [Knowledge base](https://docs.tryhamsa.com/agents/knowledge-base/overview.md).

## Good to know

* Settings changes go to your draft. Calls in progress and callers on the live version aren't affected until you publish.
* The agent never hangs up because of silence. Only the caller, smart call end, an End call step (flows) or the maximum call length end a call.
* Browser test calls use these settings too, with two differences: the **VAD activation threshold** is scaled differently, and transfers at the call limit need a phone call.

## Related

> [card] [Voice, language and dialect](https://docs.tryhamsa.com/agents/settings/voice-and-language.md)
>
> Voice, speech recognition and the dialect switcher.

> [card] [Choose the model](https://docs.tryhamsa.com/agents/settings/model.md)
>
> The language model behind the agent.

> [card] [Test your agent](https://docs.tryhamsa.com/agents/release/test-your-agent.md)
>
> Hear the effect of each change.

> [card] [Change agent settings step](https://docs.tryhamsa.com/agents/flow/steps/change-settings.md)
>
> Change settings partway through a flow.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.