How Conversation Memory works
-
Capture signals from every conversation, across every channel.
Voice, SMS, chat, and more. Conversation Memory processes each interaction to extract preferences, behaviors, and context that matter for customer conversations.
-
Continuously refine what you know.
New memories are reconciled against existing ones,so agents always see the latest, most accurate view of the customer profile.
-
Surface the right context for the next interaction.
The Recall API uses semantic search to deliver the most relevant memories, summaries, and traits, whether the next agent is human or AI.
-
Ground every response in trusted business knowledge.
Connect to Enterprise Knowledge to index FAQs, policies, product docs, web pages, and more for fast, accurate retrieval.
How Enterprise Knowledge works
-
Index the content that drives every interaction.
FAQs, policies, product docs, and web pages are processed into searchable chunks, so agents always have access to the right information.
-
Retrieve what's relevant, not everything at once.
Intelligent search surfaces only the most accurate, applicable content for each question, keeping responses grounded in facts your team has approved.
Create continuity for every customer
Ensure every interaction builds on the last, so customers don’t have to repeat themselves.
-
Start every interaction with full context
Surface what was discussed, what was promised, and what still needs resolution.
-
Reduce handle time and customer frustration
Skip repetitive questions and move directly to solving the issue.
-
Carry context across voice, chat, SMS, and more
Ensure every interaction reflects the customer’s full history with you.
Conversation Memory features
-
Observation extraction and profiles
Extracts observations from every conversation and builds a unified customer profile with identity resolution—so agents always have an accurate, evolving view of each customer across channels.
-
Recall API
Uses semantic search across observations, summaries, and traits to surface only the most relevant context – reducing noise, token usage, and errors.
-
Enterprise Knowledge
Indexes policies, FAQs, product docs, and more, so agents respond with verified business facts.
-
Summaries and traits
Capture key conversation outcomes and combine them with structured attributes like tier or location, so that agents see both the story and the data.
-
Memory controls and governance
Define what gets remembered, how it’s stored, and who can access it—with built-in controls for filtering, partitioning, deletion, and traceability.
-
Model portability and integrations
Store memory independently of any AI runtime and connect to systems like Segment, Snowflake, and Salesforce, so your data stays flexible and enriched.
Make every conversation count
Capture, unify, and activate customer memory with a complete system.
Conversation Memory FAQs
AI agent memory allows systems to retain and use information from past interactions, so conversations can build on prior context instead of starting from scratch. Twilio Conversation Memory captures what was learned from each interaction and makes it available to both human and AI agents.
Vector databases and RAG help retrieve information, but they don’t manage the full memory lifecycle. Conversation Memory goes further by extracting observations from conversations, building unified customer profiles, reconciling new information over time, and returning only the most relevant context for each interaction.
Conversation Memory continuously reconciles new observations against existing data, so the latest and most accurate view of the customer is always available. This prevents conflicting or outdated information from accumulating.
Yes. Conversation Memory is model-agnostic and works independently of any AI runtime, so you can use it with the models and tools that best fit your stack without losing memory.
Conversation Memory creates a shared layer of customer context that can be accessed by both human agents and AI agents. This ensures consistency across every interaction, regardless of who—or what—is handling it.
Conversation Memory integrates with systems like Segment, Snowflake, and Salesforce to bring in existing customer data. It also uses Enterprise Knowledge to incorporate policies, FAQs, and product documentation.
The Recall API returns only the most relevant context for each interaction using semantic search. This reduces prompt size, lowers token usage, and improves response accuracy and speed.
You can get started by exploring the APIs and documentation or reaching out to Twilio to request access and discuss your use case.
Enterprise Knowledge is a separate Twilio product that indexes your business content, including FAQs, policies, product documentation, and web pages, into a searchable knowledge base. Agents retrieve accurate, approved information on demand rather than relying on what an AI model was trained to approximate.
Conversation Memory captures and surfaces what you know about your customers: their history, preferences, and context from past interactions. It also captures and surfaces what your business knows about itself: your policies, products, and procedures. They serve different purposes and can be used independently, but work well together when agents need both customer context and verified business facts in the same interaction.
Enterprise Knowledge supports FAQs, policy documents, product documentation, raw text, and web pages. Content is indexed into searchable chunks that agents can retrieve during interactions. You can update sources at any time, and agents immediately have access to the latest version without any retraining required.
Conversation Memory is part of Twilio's agent infrastructure platform and works natively with Conversation Orchestrator and Conversation Intelligence. Orchestrator connects channels into one continuous conversation and feeds each interaction into Memory, so context carries forward as customers move across voice, messaging, and chat. Intelligence uses Memory to ground its real-time signals, like sentiment and churn risk, in verified customer history and business knowledge. Conversation Memory can also be used independently, so teams can adopt it without changing the rest of their stack.