Additional Block Agent Configurations

Along with configuring your core Block Agent components (Blocks, Turns, Edges, Conditions, and Tools), you can also customize a few other agent elements to shape your agent's behavior.

 

LLM Configuration

Customize the Large Language Model (LLM) behavior.

 

LLM Properties

Property	Description	Recommended Value
temperature	Control the randomness in your agent's responses. 0 = deterministic (predicable and factual), 1=creative	Use 0.0 - 0.4 for task-focused agents.
top_p	Nucleus sampling threshold.	Use 0.0 when using a low temperature.
frequency_penalty	Discourage word repetition.	Use 1.5 for natural variety.
presence_penalty	Encourage new topics/words.	Use 1.0 for conversational agents.
max_tokens	Maximum response length.	Use 200 for voice-first agents.
request_timeout	Seconds before timeout.	Use 5.0 for responsive interactions.
max_retries	Retry attempts on failure.	Use 3 for reliability.

 

Sample LLM Configuration

llm_settings:
	temperature: 0.0
	top_p: 0.0
	frequency_penalty: 1.5
	presence_penalty: 1.0
	max_tokens: 200.0
	request_timeout: 5.0
	max_retries: 3

 

Personality Configuration

Your Block Agent's personality defines how the agent communicates across interactions. It provides persistent context, ensuring consistent behavior regardless of which Block or Turn is active.

You can add this information via the Personality section of the Behavior & Settings tab when you create your Block Agent in the AI Studio.

 

Sample Personality Configuration

personality: |
	Your name is Sofia, you are a virtual scheduling assistant for a dental services office.
  
	MULTILINGUAL CAPABILITIES:
	- You are fully bilingual in English and Spanish
	- Detect the language AND their intent simultaneously
  
	CORE PERSONALITY AND COMMUNICATION STYLE:
	- Keep responses brief and direct (1-2 sentences typical, 3 max)
	- Avoid pleasantries, filler phrases, or restating what user said
	- On voice calls, shorter is better
  
	SYNTAX:
	- All dates should be written in Month Day Year format as text
	- Phone numbers should be written out as words in groupings
	- Never use markdown, emojis, or asterisks

 

Knowledge Collection Configuration

You have the option to include Retrieval-Augmented Generation (RAG) integration in your Block Agent to answer user questions via a knowledge base.

Note: Prior to including your Knowledge Collection in the Block Agent configuration, make sure your collection was previously created and has documentation added to it. Check out this page for more information.

 

Knowledge Collection Properties

Property	Description
context_source_collection	The Knowledge Collection name.
context_threshold	How closely aligned a match should be for the agent to identify it as the right answer to the user message.
answers_count_threshold	How many similar documents the agent should find (matching the Context Threshold) before providing an answer.
embedding_model	The model you assigned to your Knowledge Collection (always "intelepeer: text-embedding-ada-002:1536")
storage_version	Always "3". This corresponds to the database where your collection is stored behind the scenes.

 

Sample Knowledge Collection Configuration

knowledge_collections:
	- context_source_collection: "faq-tend-phase-1"
	  context_threshold: 0.8
	  answers_count_threshold: 3
	  embedding_model: "intelepeer:text-embedding-ada-002:1536"
	  storage_version: 3