Introduction. A Twilio developer tutorial published 14 August 2026 explains how to build an AI assistant that maintains persistent customer memory by integrating Twilio Agent Connect (TAC) with Twilio Flex and Twilio Conversational Intelligence, and by writing extracted preferences into Twilio Conversation Memory via the Twilio Memory API.
Key Details
The tutorial describes an end-to-end flow in which: (1) a customer contacts a Twilio number and TAC routes them to an AI assistant; (2) the assistant handles the conversation and, if needed, TAC triggers a Twilio Studio handoff flow to a Flex agent; (3) Conversational Intelligence generates a plain‑language summary at conversation end; (4) a webhook fires a Twilio Function that calls OpenAI, extracts structured preferences from the summary, and writes those values as traits into the customer’s Memory profile using the Twilio Memory API; and (5) on later contacts TAC fetches the Memory profile at session start and injects those traits into the system prompt so the assistant can respond in the customer’s preferred language automatically.
Prerequisites listed in the tutorial include Python 3.10 or newer, an active Twilio account with an SMS‑ and voice‑capable phone number, a configured Twilio Flex instance, an OpenAI account and API key, a Twilio Functions service, ngrok to expose localhost to Twilio webhooks, and a phone for testing.
Key configuration steps shown in the guide include creating a Conversation Orchestrator configuration (with Conversation memory enabled and both Observations and Summaries toggled on), copying the Conversation configuration ID and Memory SID into a local .env file, creating two Conversation Intelligence rules (one that enables Conversation Memory and a second that posts a webhook to a deployed Twilio Function), and setting up a Twilio Function (the tutorial names the service dynamictrait and a public /trait path) that invokes OpenAI to extract preferences and then patches the Twilio Memory store for the customer profile.
Who May Be Affected
This approach is relevant to organisations that use Twilio infrastructure: those operating Twilio phone numbers, running Twilio Flex instances, and able to deploy Twilio Functions. The tutorial explicitly requires an active Twilio account with an SMS‑ and voice‑capable phone number and a configured Flex instance, so teams already invested in Twilio products are the primary audience.
Why It Matters For Voice AI
The tutorial demonstrates a practical pattern for persistent memory in conversational voice or SMS assistants by extracting customer attributes (the demo uses traits named model, color, country and language) from conversation summaries and storing them in a Twilio Memory profile. The Python TAC backend sample in the guide shows memory retrieval and language‑preference support: when a profile trait for language exists the code appends an instruction to the assistant to switch language for further communication. The example also sets voice and SMS channels with memory_mode=”always” so memory is consulted on each session.
What To Watch Next
If you trial this pattern, follow the tutorial steps in order: enable Conversation memory in the Conversation Orchestrator and create a Memory store (copy the Memory SID into your .env), expose TAC locally on port 8000 via ngrok and set TWILIO_VOICE_PUBLIC_DOMAIN to the ngrok host, deploy the dynamictrait Twilio Function and paste its public URL into the second Conversational Intelligence rule, and create a Preferences trait group with individual traits (model, color, country, language) so the Function can write into those trait keys. The tutorial’s Function code calls OpenAI (the example uses model “gpt-4.1-mini”) to extract preferences and issues a PATCH to https://memory.twilio.com/v1/Stores/{store}/Profiles/{profile} to save traits.
Sources
Primary: Twilio blog, “Build Persistent Customer Memory with Twilio Agent Connect and Conversation Intelligence” (published 14 August 2026): https://www.twilio.com/en-us/blog/developers/tutorials/product/build-persistent-customer-memory-with-twilio-agent-connect-and-c
Additional material and code excerpts are taken from the same Twilio tutorial, including the step‑by‑step setup for Conversation Orchestrator, Conversation Intelligence rules, the dynamictrait Twilio Function and the TAC Python backend examples referenced in the guide.
AiDial analysis: This Twilio tutorial maps a clear implementation path for persisting conversational preferences inside Twilio’s own Memory store and demonstrates how those traits can be reinjected into prompts at session start to change assistant behaviour (for example, automatic language switching). For Australian businesses that already use Twilio phone numbers and Flex, it shows a way to avoid managing a separate customer‑profile database by keeping preferences within Twilio Conversation Memory. Organisations should follow their own risk, privacy and data governance processes when capturing and reusing customer attributes.
This article is general information and not legal advice.
