Frequently Asked Questions about AI Agents

Some answers to common issues when implementing Agent in production

Q: The AI agent seems to answer the question accurately but the answers are incomplete. What can I do?

A: To avoid this, you need to increase the Output Tokens from the “settings” section of your agent.

NOTE: If your agent use case requires a lot of memory/history of the previous chats to continue the conversation this can create huge context windows and you might get a quickly scaled up usage and higher billing.

Q: We are running a voice-based agent, the TTS model reads out the formatting syntax like asterisks, hyphens, etc. This is very troubling to the user. What can we do?

A: You can add a simple prompt instruction: “Don't include markdown” or if you want to be a lot more specific you can write a detailed prompt like this:

“Remember, you are a voice agent, so do not use markdown, HTML, latex or any other markup language, instead, output plain text without any formatting characters like asterisk, hyphen, bracket, hash, underscore etc. Keep your responses concise and clear under 100 words.”

Q: The agent is running on Whatsapp for text and Twilio for phone audio, I want to make sure that some prompts are specified only to Twilio, I don’t want to make two agents. What can i do?

A: For example, if you want to make sure that Twilio does not use markdown but you want formatting for your Whatsapp or Web agents. You can add “if statement” to the prompt through our Jinja templating language like this:

{% if platform == "TWILIO" %}
Remember, you are a voice agent, so do not use markdown, HTML, latex or any other markup language, instead, output plain text without any formatting characters like asterisk, hyphen, bracket, hash, underscore etc. Keep your responses concise and clear under 100 words.
{% endif %}

NOTE: This can be added in your basic prompt instructions there are no “special” settings for this.

Q: I have a WhatsApp/Slack agent, but the text output shows up as markup language. How can I fix this?

A: By incorporating Jinja templating in the Prompt, you can adjust the prompt as needed. In the example below we are using WhatsApp and Slack clients running alongside your web client you can try this example in your AI agent:

You are an AI Bot that runs the front-desk of a Dentist's clinic.

{% if platform == [ "WHATSAPP", "SLACK" ] %}
    Remember, you are a {{ platform }} agent, so do not use HTML, latex or any other markup language, instead use only the following formatting styles: 
    italic: single underscore
    bold: single asterix. Do not use 2 asterix as per markdown, instead use {{ platform }} guidelines and use 1 asterix only
    strikethrough: single tilde
    code: single backtick
    code block: 3 backticks
    quoted text: place an angle bracket and space before the text
    headings: just use bold style with 1 asterix.
{% endif %}

Here the Jinja templating language uses an if statement to provide the prompt with a condition to output the text in a particular way based on the platform the agent is integrated into.

Q: How many last messages (conversation history) does the agent accept while answering the user's next question?

A: We retain up to 50 conversations in the conversation history.

Q: Can LLM "temperature" be increased beyond 1 if I want more creative responses?

A: We host several AI models, and the LLM "temperature" settings can vary. Here are the temperature settings per model:

AI Model	Temperature Settings
OpenAI	0.0 to 2.0
Claude 3.5 Sonnet	0.0 to 1.0
Gemini 1.5 Pro	0.0 to 2.0
Gemini 1.5 Flash	0.0 to 2.0
Gemini 1.0 Pro Vision	0.0 to 1.0