Adapters

tac.adapters

Adapters for integrating with external services in the Twilio Agent Connect.

AdapterOptions

Bases: BaseModel

Options for configuring how adapters inject memory and profile data.

Example

Default behavior (no options) - inject ALL profile traits

client = with_tac_memory(openai_client, memory_response, context)

Default behavior (options but no profile_traits specified) - inject ALL profile traits

options = AdapterOptions() client = with_tac_memory(openai_client, memory_response, context, options=options)

Explicitly exclude all profile traits

options = AdapterOptions(profile_traits=None) client = with_tac_memory(openai_client, memory_response, context, options=options)

or

options = AdapterOptions(profile_traits=[]) client = with_tac_memory(openai_client, memory_response, context, options=options)

Specific traits only

options = AdapterOptions(profile_traits=["Contact", "Preferences"]) client = with_tac_memory(openai_client, memory_response, context, options=options)

get_profile_traits

get_profile_traits() -> list[str] | None

Get the profile traits to include.

Returns:

Type	Description
list[str] | None	None to include all traits (when field not set),
list[str] | None	empty list to exclude all (when explicitly set to None or []),
list[str] | None	or list of specific trait group names to include.

MemoryPromptBuilder

Builds LLM prompts from TAC memory and profile data.

This class orchestrates prompt building by calling helper methods on TACMemoryResponse and ConversationSession models, then assembles the sections into a complete prompt.

Example

prompt = MemoryPromptBuilder.build(memory_response, context, options) if prompt: ... # Inject into your LLM messages ... messages.insert(0, {"role": "system", "content": prompt})

build staticmethod

build(
    memory_response: TACMemoryResponse | None = None,
    context: ConversationSession | None = None,
    options: AdapterOptions | None = None,
) -> str | None

Build a complete memory prompt from TAC data.

This is the main entry point. Delegates formatting to model helper methods, then assembles sections into a complete prompt.

Parameters:

Name	Type	Description	Default
memory_response	TACMemoryResponse | None	Memory data from TAC.retrieve_memory()	None
context	ConversationSession | None	Conversation session with profile data	None
options	AdapterOptions | None	Adapter options for trait filtering	None

Returns:

Type	Description
str | None	Formatted prompt string ready for LLM injection, or None if
str | None	no memory/profile data is available.

Example

prompt = MemoryPromptBuilder.build( ... memory_response=memory_response, ... context=context, ... options=AdapterOptions(profile_traits=["Contact"]), ... ) print(prompt)

Customer Context

You have access to the following information about this customer from previous interactions:

Customer Profile

Information about this customer: - Contact: {"name": "John Doe", "email": "john@example.com"}

Key Observations

Important notes about the customer from previous interactions: - Customer prefers email communication

compose staticmethod

compose(
    system_prompt: str | None = None,
    memory_response: TACMemoryResponse | None = None,
    context: ConversationSession | None = None,
    options: AdapterOptions | None = None,
) -> str

Compose system prompt with memory context.

Appends memory to system_prompt if available. Always returns a string.

Example

prompt = MemoryPromptBuilder.compose( ... "You are a helpful assistant", memory_response, context ... )