--- title: "Building a Multi-Channel AI Agent with Twilio Conversations and eve" description: "Building a multi-channel AI agent that remembers customers across SMS and WhatsApp is challenging because most projects require a custom memory layer; Twilio Conversations and Vercel’s eve solve this by cleanly separating short-term and long-term memory. eve holds the per-conversation transcript, while Twilio’s Memory Store persists extracted facts across sessions using automatic observation extraction from Conversation Intelligence. The tutorial walks through provisioning an Orchestrator Configuration that stitches SMS and WhatsApp into one thread, wiring a dynamic instruction that merges both memory sources, and using ngrok for local webhook tunneling. The resulting agent remembers customer context across channels and sessions, allows model swapping in one line via the AI SDK, and extends cleanly to voice and RCS without bespoke persistence code. The full sample project is on GitHub at IObert/twilio-conversations-eve-agent, and the tutorial includes provisioning steps using curl, environment variable setup, and local testing with pnpm dev." --- # Building a Multi-Channel AI Agent with Twilio Conversations and eve [Twilio](https://yomu.fyi/company/twilio) · Marius Obert · Oct 8, 2026 **Type:** Tutorial ## Summary Building a multi-channel AI agent that remembers customers across SMS and WhatsApp is challenging because most projects require a custom memory layer; Twilio Conversations and Vercel’s eve solve this by cleanly separating short-term and long-term memory. eve holds the per-conversation transcript, while Twilio’s Memory Store persists extracted facts across sessions using automatic observation extraction from Conversation Intelligence. The tutorial walks through provisioning an Orchestrator Configuration that stitches SMS and WhatsApp into one thread, wiring a dynamic instruction that merges both memory sources, and using ngrok for local webhook tunneling. The resulting agent remembers customer context across channels and sessions, allows model swapping in one line via the AI SDK, and extends cleanly to voice and RCS without bespoke persistence code. The full sample project is on GitHub at IObert/twilio-conversations-eve-agent, and the tutorial includes provisioning steps using curl, environment variable setup, and local testing with pnpm dev. ## Context Developers building AI agents for customer support often struggle with maintaining memory across different communication channels and sessions. This tutorial addresses that by combining Twilio Conversations for long-term memory and Vercel’s eve for short-term memory. ## Approach / What changed The tutorial provides a step-by-step guide to provisioning a Twilio Conversations Orchestrator Configuration, connecting SMS and WhatsApp channels, setting up a Memory Store with automatic observation extraction, and wiring an eve agent with a dynamic instruction that merges short-term transcript and long-term memory. It uses ngrok for local webhook tunneling and the AI SDK for model integration. ## Takeaways - Twilio Conversations separates memory into three products: Orchestrator for cross-channel threading, Memory Store for persistent customer profiles, and Intelligence for automatic observation extraction. - eve holds the per-conversation transcript as short-term memory, while Twilio Memory Store persists extracted facts across sessions, eliminating the need for custom persistence code. - The agent can be extended to voice and RCS by plugging into the same Orchestrator Configuration, and the model can be swapped in one line by changing the AI SDK provider. ## FAQ ### How does the agent remember customers across different channels? Twilio Conversations' Orchestrator stitches SMS and WhatsApp into one thread per customer, and the Memory Store persists extracted facts (traits and observations) across sessions. The agent injects these facts into the prompt via a dynamic instruction, so the customer's history is available on the first turn of any new conversation. ### What is the difference between eve's memory and Twilio's memory? eve holds the short-term per-conversation transcript, which is reset when a session ends. Twilio's Memory Store holds long-term facts extracted from past conversations, such as customer preferences and order details. The dynamic instruction merges both at the start of each session. ### How do I test the agent locally? Use ngrok to expose a public HTTPS URL for webhooks, run pnpm dev to start the eve agent locally, and provision the Orchestrator Configuration with the ngrok URL. Then send an SMS or WhatsApp message to your Twilio number to trigger the webhook and see the agent respond. **Tags:** [AI Agents](https://yomu.fyi/topic/ai-agents), [eve](https://yomu.fyi/topic/eve), [Memory](https://yomu.fyi/topic/memory), [Twilio](https://yomu.fyi/topic/twilio), [WhatsApp](https://yomu.fyi/topic/whatsapp) - Source: [Twilio](https://www.twilio.com/en-us/blog/developers/building-agent-twilio-conversations-eve) - Source URL: https://www.twilio.com/en-us/blog/developers/building-agent-twilio-conversations-eve - Ingested by Yomu: 2026-10-09T06:00:35.233Z [Read original post](https://www.twilio.com/en-us/blog/developers/building-agent-twilio-conversations-eve)