Twilio Agent Connect is a software development kit that links an AI agent you build to Twilio's communication channels. It became generally available on May 6, 2026, for Python and TypeScript. Your agent receives messages from voice, SMS, chat, WhatsApp and RCS through one set of callbacks, and Twilio supplies the conversation memory, orchestration and route to a human agent.
The details below come from Twilio's general availability announcement. We have not built a production agent on it, so treat this as a map of what Twilio says it does, not a review.
What problem does Agent Connect solve?
A basic SMS AI agent is simple. A webhook receives an inbound text, sends it to a model and posts the reply back through the Messaging API. Our Twilio SMS agent tutorial walks through that pattern.
The hard parts show up later:
- Memory. SMS is stateless. Every inbound message arrives on its own, so you need to store and retrieve the conversation yourself.
- More channels. Once the agent works on SMS, someone asks for WhatsApp or voice. Each channel has its own payload format and quirks.
- Handoff. Some conversations need a person, and the person needs the history.
Agent Connect is Twilio's answer to those three. It sits between your agent code and Twilio's channels, normalises the events and connects to Twilio's memory and contact-centre products.
- Customer on SMS, WhatsApp, RCS or voice
- Twilio channels
- Agent Connect callbacks
- Your agent and model
- Reply or Flex handoff
What does Agent Connect include?
According to Twilio's announcement, the general availability release includes:
| Component | What Twilio says it does |
|---|---|
| Unified channel callbacks | One interface for voice, SMS, chat, WhatsApp and RCS |
| Conversation Orchestrator integration | Recommended for multi-channel agents |
| Conversation Memory | Semantic recall, customer profiles and conversation history |
| Lifecycle hooks | Points where your code runs as a conversation progresses |
| LLM adapters | Starting with an adapter for the OpenAI SDK |
| Built-in tools | Memory retrieval, knowledge search and escalation to Twilio Flex, compatible with OpenAI and Anthropic function calling |
| Cloud connectors | Azure (Agent Framework, Voice Live) and AWS (Strands, Bedrock, AgentCore) |
| TACServer | A server built on FastAPI for Python or Fastify for TypeScript |
| ConversationRelay-only mode | For voice-first agents |
| Conversation Intelligence | Optional |
- Agent Connect generally available
- May 6, 2026
- SDK languages: Python and TypeScript
- 2
- Channels: voice, SMS, chat, WhatsApp, RCS
- 5
- Listed in the GA announcement
- No price
It shipped alongside two related pieces. Conversation Orchestrator became generally available on May 5, 2026, and Conversation Memory keeps context across SMS, WhatsApp, RCS and voice. Our platform comparison sets these against Sinch, Infobip and Bird.
How does it fit into an SMS agent architecture?
In the reference design we describe in SMS agent architecture, an SMS agent has five layers: a webhook that validates the provider's signature, a compliance gate that handles STOP and HELP before any AI runs, a state store keyed by phone number, an agent loop that calls the model and approved tools, and an outbound queue with a human handoff path. Agent Connect covers parts of the webhook, state and handoff layers. The compliance gate is still your job.
You still own:
- The model and prompt. Agent Connect is bring-your-own-agent. Twilio supplies adapters, not the model.
- Business tools. Booking calendars, CRMs and order systems are yours to connect.
- Compliance. 10DLC or toll-free registration, consent and opt-out handling still apply. See our 10DLC registration guide.
- Safety. Inbound texts are untrusted input. A memory layer that recalls old messages can also recall a planted instruction. Our guide to prompt injection via messages covers defences.
- Register your numberComplete 10DLC or toll-free verification before sending.
- Install the SDKUse the Python or TypeScript package and run the TACServer.
- Connect your modelUse the OpenAI SDK adapter or a cloud connector for Azure or AWS.
- Add toolsEnable memory retrieval and knowledge search, and add your own business functions.
- Set escalationRoute conversations that need a person to Twilio Flex with history attached.
How much does Agent Connect cost?
Twilio's general availability announcement does not list a price for Agent Connect itself, as of November 2026. The services it uses are priced separately on Twilio's public pages:
- US SMS: $0.0083 per segment in either direction, plus carrier fees.
- Conversation Orchestrator: $0.0002 per 1,000 characters.
- ConversationRelay (voice AI): $0.07 per minute plus voice charges.
We did not find published prices for Conversation Memory or Conversation Intelligence in the sources we checked. Ask Twilio for a quote before you model costs at scale, and add your own LLM costs on top. Our SMS AI agent cost guide and 10DLC fee breakdown cover the rest of the bill.
When should you use Agent Connect, and when not?
Pros
- One agent across SMS, WhatsApp, RCS, chat and voice
- Memory and Flex escalation without building them yourself
- Works with OpenAI, Anthropic function calling, Azure and AWS frameworks
- Python and TypeScript with a ready server
Cons
- Ties memory and orchestration to Twilio
- No published price for the SDK or memory layer
- More moving parts than a single webhook
Agent Connect makes the most sense if you plan to run one agent across several channels, already use Twilio Flex for human agents, or want Twilio-managed memory instead of your own database.
A plain webhook is probably enough if you only need SMS, the conversations are short (appointment reminders, simple FAQs), or you want to keep conversation data in your own systems. For a voice-only build, our ConversationRelay tutorial covers the voice layer directly.
How does it relate to MCP?
They solve different problems. Agent Connect connects an agent to customer channels. MCP connects an agent, or a coding assistant, to tools and data. Twilio also publishes an MCP server that gives coding agents search over its API documentation, which helps when you write the integration. See SMS agents and MCP, and for the general idea, getpersonalaiagents.com on MCP connectors.
Bottom line
Agent Connect is Twilio's SDK for connecting your own AI agent to SMS, WhatsApp, RCS, chat and voice, with memory and human handoff built in. It has been generally available since May 6, 2026, in Python and TypeScript. It saves real work if you need several channels or Flex escalation, but it adds dependency on Twilio and has no published standalone price. For a single-channel SMS bot, start with a webhook and move to Agent Connect when you need what it adds.


