Skip to main content

Quickly Configure an AI Customer Service Phone Bot

This article will guide you through quickly configuring a phone-based AI customer service bot that can hold conversations.For simplicity, this example uses openrouter.ai as the LLM provider, and Deepgram for speech recognition and text-to-speech.

1. Apply for API Keys

Apply for an API key for the LLM.

You need to register an account on openrouter.ai and add funds to your account.
After the top-up is completed, create an API key in the OpenRouter dashboard and record its value. It is recommended to save it in a text file because it will be used later.

Apply for a Deepgram API key.

Go to deepgram.com and register an account. After registration, the account includes $200 in free trial credits by default.
Create an API key in the Deepgram dashboard and record its value.

2. LLM Configuration

Admin menu: Basic Configuration -> Large Language Models , then click Add.
The system includes a built-in OpenRouterChat example for your reference. 
Pay attention to the following fields:

Max Concurrency: Set this to 0, which means unlimited.
Server URL: https://openrouter.ai/api/v1/chat/completions
Model Name: qwen/qwen3.5-flash-02-23

Available models can be found on the https://openrouter.ai/models page.
Model-related parameters can be found in the OpenRouter dashboard.
Note: Phone call scenarios are sensitive to latency, so it is recommended to choose a model with flash in its name.

image.png

image.png

image.png

3. Modify Text-to-Speech Settings

Admin menu: Basic Configuration -> Text to Speech -> Deepgram .
Modify the API Key, then save.

4. Modify Speech Recognition Settings

Admin menu: Basic Configuration -> Speech to Text -> Deepgram .
Modify the API Key, then save.

5. Inbound Customer Service Configuration

Admin menu: Call Center -> Inbound Routing , then click Add.
The system supports configuring multiple inbound numbers. Different inbound numbers can use different LLM providers.
For both text-to-speech and speech recognition settings here, select Deepgram.

image.png

6. Test the Inbound Customer Service Bot

Before testing, you need to install a softphone.
MicroSIP is recommended. Download address: microsip.org/downloads.

Open the softphone, register the extension, and test the call:

Extension number and password: 15005600327 / 150056003027
Here, the extension number and password can be entered freely.  It is recommended to use your own mobile phone number.
Registration address: 192.168.67.230:5080 .  
Please replace 192.168.67.230 with your server address.

After registration succeeds, dial: 76543210.

image.png

image.png

You can then test inbound calls to the AI customer service bot.
In the Inbound Call Records menu, you can view conversation records and listen to call recordings.

If everything is working properly, you should be able to hear the voice prompt at this point.
If you cannot hear the voice prompt, check the call record and review the hangup cause.
For more complex issues, you may need to check the easycallcenter365 and FreeSWITCH logs.

View the easycallcenter365 log:

/home/easycallcenter365/logs/easycallcenter365-ERROR.log

View the FreeSWITCH log:

/home/freeswitch/var/log/freeswitch/freeswitch.log