Build omnichannel AI agents that talk, text, and email

A customer calls. Your voice agent picks up, handles the request, and ends the call cleanly. Good.

Then what?

The confirmation goes out by email. The delivery update goes by SMS. The support thread continues on WhatsApp because that is where the customer already lives. Three channels, one customer, one conversation and, in most stacks, three vendors, three API keys, three sets of webhooks, and three places where the context quietly falls on the floor.

An agent that cannot see what happened on the other channels is an agent that repeats itself, contradicts itself, and asks the customer to explain something they already explained twice.

With Email API, Telnyx covers the last channel that used to require a second vendor. Voice, SMS, MMS, WhatsApp, fax, and email on one platform, one API key, one event stream, one bill.

Omnichannel is not "we support a lot of channels

Plenty of vendors will sell you a list of channels. That is not the same thing.

Omnichannel means the conversation is the unit, not the message. The customer does not think of "the call" and "the email" as separate events. They think of it as the time they dealt with your company. When your systems treat those as unrelated records in unrelated vendors, the customer feels it immediately.

What makes it real:

Channels are outputs, not integrations

This is the shift that matters for anyone building with LLMs right now.

In the old model, each channel was a project. Adding email meant a vendor evaluation, a contract, an SDK, a new set of credentials, a new webhook handler, a new dashboard, a new on-call runbook. Channels were expensive enough that you picked two and lived with the consequences.

In the agent model, a channel is just a tool your model can call. Send an email. Send a text. Place a call. Check whether the last message landed. The model decides which one fits the moment, and the fit is usually obvious.

Every one of those is a function call away from the same API key. The agent that just finished the call can send the invoice before the customer has put the phone down.

Connect your LLM to any channel

You do not need to hand-write integrations to get there. The Telnyx AI repo is the official one-stop shop for building agents on Telnyx, open source under MIT.

Agent Toolkit

Telnyx APIs exposed as function-calling tools for the frameworks you already use — OpenAI's Agents SDK, LangChain, CrewAI, and Vercel's AI SDK. Available in Python (telnyx-agent-toolkit) and TypeScript (@telnyx/agent-toolkit). You configure which actions the agent is allowed to take, and the toolkit hands your framework the tool definitions.

MCP server

A hosted Model Context Protocol server at https://api.telnyx.com/v2/mcp, or run it locally with npx @telnyx/mcp. Any MCP-capable client gets Telnyx as a first-class tool provider. Focused MCP Apps cover specific workflows: number intelligence, usage and cost exploration, voice monitoring.

Agent Skills

228 and counting, covering messaging, voice, numbers, AI, IoT, WebRTC, and Twilio migration. Skills give coding agents accurate, current context about how these APIs actually work, instead of letting them guess from training data that predates the endpoint they are calling.

/plugin marketplace add team-telnyx/ai
/plugin install telnyx-email@telnyx

Agent CLI

Composite commands that collapse multi-step provisioning into one line — telnyx-agent setup-sms, setup-voice, setup-ai. Every command supports --json, because the thing running it is usually not a human.

Harness plugins

Ready-made providers for OpenClaw and Hermes, covering voice calls, TTS, STT, embeddings, inference, and SMS channels.

The point of all of it is the same: your model should be able to reach a customer on any channel without you writing a bespoke integration first.

What consolidation actually buys you

Not "fewer logos on the architecture diagram." Concrete things.

Context that survives the channel switch. Delivery, open, and reply events from every channel land in one place. The agent handling the follow-up knows the receipt bounced before the customer calls to say they never got it.

A cost model you can reason about. Pay-as-you-go across channels on one invoice. Comparing what a resolution costs by voice versus by message stops being a spreadsheet exercise.

One vendor relationship. One contract, one renewal, one support queue, one compliance review. Anyone who has run a procurement cycle for a third messaging vendor knows what that is worth.

Carrier-owned infrastructure underneath. Telnyx owns the network the voice and messaging run on. Email runs on the same platform, on the same MTA stack, not a reseller layer with someone else's reputation attached.

Room to add the next channel. When the channel your customers move to next arrives, it is a new tool in the same agent, not a new vendor evaluation.

Where to start

Email is the piece most teams are still outsourcing. It is also the piece that closes the loop on almost every agent workflow: the receipt after the call, the summary after the chat, the document that has to exist somewhere the customer can find it later.

• Quickstart — set up a sending domain and send your first message
• API reference — full endpoint documentation
• Support — guides and answers from the team
• github.com/team-telnyx/ai — toolkits, MCP, skills, and CLI for building the agent that uses all of it