Skip to main content
This guide takes you from an empty project to an agent that people can call. You create a single prompt agent, write a short prompt, pick a voice, talk to it in your browser, publish it, and then put it on your website or a phone number.

Before you start

  • A Hamsa account. Every account comes with a Default project and free credits to try Hamsa. Calls use credits, and a call can only start while your balance is above zero.
  • The Owner or Admin role in the project, to publish. A User can do every step up to publishing, then asks an Owner or Admin to merge their draft (see Merge requests).
  • A computer with a microphone, and a browser that is allowed to use it.
  • For the phone part: a phone number from Twilio or a SIP provider.

1. Create the agent

1

Open the Agents list

In the sidebar, click Agents list.
2

Create a single prompt agent

Click Create agent and choose Create prompt agent.
The Create agent menu open on the Agents list, with Create prompt agent and Create flow agentThe Create agent menu open on the Agents list, with Create prompt agent and Create flow agent
3

Name it

Enter an Agent name, for example “Acme front desk”, and click Create. The single prompt editor opens.
The Create single prompt agent dialog with Acme front desk typed in Agent name and the Create buttonThe Create single prompt agent dialog with Acme front desk typed in Agent name and the Create button
The editor has two cards on the left, Greeting message and Agent instructions, and the Configuration panel on the right. Your edits save on their own to your draft, which only you see. See Single prompt agents for everything on this screen.
The single prompt editor: Greeting message and Agent instructions on the left, the Configuration panel with its sections on the right, and Test and Publish live in the headerThe single prompt editor: Greeting message and Agent instructions on the left, the Configuration panel with its sections on the right, and Test and Publish live in the header

2. Write the greeting and the prompt

1

Write the greeting

On the Greeting message card, keep the Static message tab and replace the text with your first line, for example “Hello, this is Sara from Acme Dental. How can I help you today?” The greeting can be up to 191 characters.
2

Write the instructions

In Agent instructions, replace the starting text with a short prompt. Say who the agent is, what it does, the rules it follows and the facts it needs:
3

Leave the prompt enhancer on Basic

Under the instructions, Prompt enhancer is set to Basic. It adds Hamsa’s speaking rules and tells the model to speak the voice’s dialect, so you don’t need to write those yourself.
The Prompt enhancer list open under the instructions, showing Disabled, Basic (selected) and AdvancedThe Prompt enhancer list open under the instructions, showing Disabled, Basic (selected) and Advanced
For what makes a prompt work on a live call, see Write a prompt that works on calls.

3. Pick a voice

The voice decides how the agent sounds and which language and dialect it speaks. New agents start with Carla, an Arabic voice.
1

Open the voice picker

In Configuration, open Voice settings and click Change on the voice row. Select agent voice opens.
2

Find a voice in your callers' dialect

Filter by Dialect and Gender, and press play on a row to hear it.
Select agent voice in the Configuration panel, with the Dialect filter open listing Gulf and Levantine dialects and how many voices each hasSelect agent voice in the Configuration panel, with the Dialect filter open listing Gulf and Levantine dialects and how many voices each has
3

Pick it

Click the voice. The picker closes and the agent now speaks that voice’s language and dialect.
If your prompt or greeting is in another language than the voice, change them to match. See Voice, language and dialect.

4. Check two call settings

Open Call settings in Configuration:
Call settings in the Configuration panel: Wait for user to speak first, On-hold music, Timezone, User gender detection and Smart call endCall settings in the Configuration panel: Wait for user to speak first, On-hold music, Timezone, User gender detection and Smart call end
  • Smart call end: turn it on so the agent can say goodbye and hang up when the conversation is done. With it off, a single prompt agent never hangs up on its own.
  • Timezone: pick your business’s time zone, so the agent gets dates and times right.
Every call is also capped by Max call duration, 5 minutes for a new agent. Raise it if your calls run longer. See Call behavior settings.

5. Test it in the browser

1

Start a browser call

In the editor header, click Test and choose Browser call. If the browser asks for the microphone, allow it. The Test button is disabled while changes are still saving.
The Test menu open in the editor header, with Browser call, Phone call, Chat and Test valuesThe Test menu open in the editor header, with Browser call, Phone call, Chat and Test values
2

Talk to the agent

Ask what a caller would ask. Then try the cases your rules cover: an off-topic question, a request for medical advice, a name the agent should repeat back. While the call runs, the Configuration panel shows Live call logs.
3

End the call

Click End call. You can also start and end a browser call with Cmd+Shift+Enter on Mac or Ctrl+Shift+Enter on Windows.
4

Read the call back

In the editor header, open More actions and click Calls, then open your test call. Read the transcript, then fix the prompt where the agent hesitated or broke a rule, and test again.
Tests run your own draft. Nobody else hears your changes until you publish. See Test your agent.
The single prompt editor during a browser test call: End call and the microphone button in the header, and Live call logs on the right listing the call start, the greeting, model replies and a tool callThe single prompt editor during a browser test call: End call and the microphone button in the header, and Live call logs on the right listing the call start, the greeting, model replies and a tool call

6. Publish

Callers only ever hear a published version. Publishing turns your draft into version v1.
1

Click Publish live

In the editor header, click Publish live.
2

Say what changed

Fill in Change summary, for example “First version of the front desk agent”. It is required. New version label is optional.
The Publish live dialog with a change summary filled in, an empty New version label and the Publish live buttonThe Publish live dialog with a change summary filled in, an empty New version label and the Publish live button
3

Confirm

Confirm to publish. Callers hear the new version from the next call.
If you are a User, click Request merge instead, and an Owner or Admin publishes it. To test in dev and staging before prod, turn on deployments later; see Environments and releases.

7. Put the agent in front of callers

Choose one, or do both.
1

Open the publish page

In the agent editor, open More actions (the three dots) and click Publish page. Publish agent opens with a live preview.
2

Set the look

Pick a Theme and an Audio visualizer, and under Embed set the Position and Button label of the floating button. Everything is optional.
3

Publish the page

Click Publish agent. You get a Public link that anyone can open to talk to the agent, with no sign-in.
4

Add the widget to your site

Turn on the Embed switch, click Copy, and paste the snippet before </body> on each page where the button should appear. Your site must be served over HTTPS for the microphone to work.
The Embed section of the Publish agent page, with Position, Size, Button label and Auto-start call, a preview of the floating button and the Publish agent buttonThe Embed section of the Publish agent page, with Position, Size, Button label and Auto-start call, a preview of the floating button and the Publish agent button
Visitors talk to the published version. See Embed on your website for widget options and the web SDK.

8. Watch the calls

Every call, including your tests, is in the Conversation log in the sidebar, with its transcript, recording and an analysis of how it went. To get each call’s results in your own system as it ends, set a webhook.

Next steps

Add a knowledge base

Give the agent documents and text to answer from.

Connect an API

Let the agent look up and change things in your systems.

Build a flow agent

For calls with fixed stages and rules.

How it fits together

The ideas behind projects, versions and environments.