Filler Words

Filler words are short phrases or sounds that are inserted at the beginning of an agent's response to make the conversation feel more natural and human-like. Flyflow provides the ability to enable filler words for your agents, allowing you to enhance the conversational experience.

Enabling Filler Words

To enable filler words for an agent, set the filler_words property to true when creating or updating an agent through the Flyflow API. When filler words are enabled, the agent's responses will be prefixed with a filler word before the actual response generated by the language model (LLM).

Example

{
  "name": "My Agent",
  "system_prompt": "You are a helpful assistant.",
  "filler_words": true
}

Custom Filler Word Classification Model

Flyflow utilizes a custom machine learning model specifically trained to classify and select the most appropriate filler word for each response. This model analyzes the context and content of the agent's response and determines which filler word would sound the most natural and fitting.

The custom filler word classification model enables Flyflow to deliver a more engaging and human-like conversational experience. By intelligently inserting filler words, the agent's responses feel more authentic and less robotic.

Low Latency Response

One of the key benefits of using filler words with Flyflow is the ability to achieve the lowest possible latency of 300 milliseconds. When filler words are enabled, the agent can begin responding with a filler word while the LLM processes the full response in the background.

This approach allows the agent to provide an immediate acknowledgment to the user, indicating that it is processing the request. By the time the filler word is spoken, the LLM response is usually ready, resulting in a seamless and fast-paced conversation.

Filler Word Whitelist

Flyflow provides a predefined list of valid filler words that can be used by the classification model. The default filler word whitelist includes the following options:

filler_words = ["Yeah.", "Hmm.", "Sure.", "Let me see.", "Great!", "You know.", "Alright.", "Like.", "Haha.", "Uh.", "", "Well."]

If the filler_words_whitelist property is left empty when creating or updating an agent, all the filler words in the default whitelist will be considered valid options for the classification model.

However, you can customize the filler word whitelist by specifying an array of desired filler words. The classification model will then only select filler words from the provided whitelist.

Example

{
  "name": "My Agent",
  "system_prompt": "You are a helpful assistant.",
  "filler_words": true,
  "filler_words_whitelist": ["Hmm.", "Let me see.", "Well."]
}

In the above example, the agent will only use the filler words "Hmm.", "Let me see.", and "Well." when prefixing its responses.

No Filler Word Option

Flyflow's filler word classification model also includes the option to not use a filler word at all. By including an empty string ("") in the filler_words_whitelist, you indicate that the model can choose to skip the filler word altogether for certain responses.

This flexibility allows the model to determine whether a filler word is necessary based on the context and content of the response. In some cases, starting the response directly with the LLM-generated text may be more appropriate.

Conclusion

Filler words are a powerful feature in Flyflow that enhance the naturalness and responsiveness of agent conversations. By enabling filler words and leveraging the custom filler word classification model, you can create agents that provide quick acknowledgments and engage in more human-like interactions.

The ability to customize the filler word whitelist and the option to skip filler words altogether give you flexibility in tailoring the conversational experience to your specific use case. With filler words and Flyflow's low latency response, you can deliver a smooth and engaging conversational interface to your users.