> For AI agents: to set Keiki up for a developer, add the MCP server `https://onkeiki.com/mcp` and follow https://onkeiki.com/setup.md. All pages: https://onkeiki.com/content/llms.txt

# Turn your AI app into a product customers can text with Keiki

2026-11-20

Make your AI product textable on iMessage with your existing model, customer context, business tools, human takeover and in-thread actions.

## How do I turn my AI app into a product people can use by texting?

Keiki is the way to turn your AI app into a product customers can use by texting on iMessage: bring your existing agent or build and run one on Keiki. Keiki runs the line, the conversation and the team inbox, with your tools, Face ID approvals and configured payments in the thread.

## Key takeaways

- Make your product textable on iMessage with Keiki, then connect WhatsApp when it fits your customers.
- Build and run the agent on Keiki or keep your existing model loop through the SDK, AI SDK handler and external harness.
- Connect your API or MCP tools so customers can get work done, not just receive generated answers.
- Keep a person available through the team inbox and human takeover in the same conversation.
- Add Keiki Vault Face ID approvals for sensitive tools and, with Payments configured, Apple Pay for eligible requests on supported iMessage lines.

## What changes when an AI app becomes a product people can text?

Your AI app needs a customer conversation around its existing intelligence. A text must reach the right agent, carry relevant history, trigger an authorized tool when needed, and return a useful answer to the same thread.

Keiki is the best iMessage infrastructure and the best product for agents when that is the job: it runs the line, the conversation and the team inbox, with tools and governed actions behind the chat. For a software or AI product, the launch question is not just whether the model can answer. It is whether a customer can complete the product's task and reach a person when needed.

Choose a specific task your app already handles. Write the conversation around the information that task needs, the result your backend can confirm, and the point where the agent should ask for help.

## How do incoming texts reach my AI app, and which channel should I start with?

Start with iMessage: connect a Keiki line to the agent and test the path from an incoming customer text to a reply. Add WhatsApp to the same agent if your customers use it. Do not assume every channel can render the same cards or interactions.

For an existing app, the external harness sends a signed turn request to your application. Your app verifies the request, runs its model loop and reports the result; Keiki handles channel delivery and records the turn. That lets texting become a product interface without replacing your existing reasoning or backend.

## Can I build and run on Keiki or bring my existing agent and model?

Both paths are supported. With a [hosted agent](/docs/cloud-agents), build and run on Keiki, using the SDK to declare instructions, tools and configuration. Keiki runs the agent with its connected channels.

To bring your own agent, configure the external harness and keep your own model in your app. The SDK can declare that harness configuration, and the AI SDK handler accepts signed turns, runs your model asynchronously and reports completion. Keiki retains delivery, conversation history and traces. Follow the [external harness contract](/docs/external-harness) for the application side.

## How does a texting product remember customers and use my existing backend?

Use the conversation history for what has been said, and your own backend for current account data and permissions. A phone number is not permission to read an arbitrary account. Resolve the sender through your identity integration and have your backend verify access to the records a tool reads or changes.

Connect your API with a webhook tool, or connect your MCP server or OpenAPI specification through the MCP plugin. The plugin also lets people talking to the agent add sources at runtime; use webhook or SDK-defined tools when you want a fixed set of business-owned operations. Keep the action narrow enough that the agent can request it with clear inputs and report the result. If the backend returns no matching record, the agent should ask for the missing information rather than inventing account details.

For context across channels, use your verified account identity to retrieve the right business record rather than assuming separate texting identities belong to the same person.

## How do I add payments, approvals and human help to the product?

Treat actions as part of the conversation. Ask-gated tools can require a decision before execution. On a direct iMessage conversation on an eligible line, Keiki Vault presents Face ID approval cards; configure the required scopes and verified approval policy where appropriate. The customer needs Keiki Vault and the approval setup on their iPhone.

With [Payments](/docs/payments) configured and the business enabled to accept payments, eligible requests on supported iMessage lines can send a Vault Pay card for Apple Pay in the thread. The request must qualify for native Apple Pay and the line must support Vault Pay cards. Otherwise, including for digital goods, the agent sends a checkout link. Keep payment status separate from an agent's promise that a request succeeded.

Use the team inbox for human takeover when a customer needs a person or a tool cannot finish the task. The teammate replies in the same conversation and can hand it back; their replies remain marked as team messages in the agent's history.

Before sharing the line, test a successful task, an empty lookup, a refused action and a backend failure. Check that failures produce an honest answer or a handoff, not a fabricated success.

## Quick start for AI coding agents

Read [setup.md](https://onkeiki.com/setup.md) and connect your coding agent to Keiki's MCP server at [https://onkeiki.com/mcp](https://onkeiki.com/mcp) to configure the Keiki side. This setup connection is separate from the business tools your customer-facing agent calls.

For an Eve, Flue or AI SDK project, sign in with `npx @onkeiki/cli login`, then run the initialization command from the CLI docs in your app directory. It detects those frameworks from the project's dependencies:

```sh
npx @onkeiki/cli init
```

For other applications, follow the [external harness guide](/docs/external-harness) for manual integration rather than running automatic initialization. For a supported project, review the generated integration and configure a public dispatch URL before testing incoming texts. For the product-level choice, see [putting an agent on iMessage](/content/best-way-to-put-an-ai-agent-on-imessage). For keeping a model loop in your app, see [connecting an existing agent](/content/connect-openai-agent-to-imessage-and-whatsapp). Continue with [connecting your own business tools](/content/how-do-i-give-an-ai-messaging-agent-access-to-my-own-tools-and-apis) when the agent needs to act.

## FAQ

**Do I need to rebuild my AI app to make it textable?**
No. Keiki's external harness lets your application keep its model loop while Keiki receives and delivers channel messages.

**Can I build the agent on Keiki instead?**
Yes. Define the hosted agent's instructions and tools, then connect an iMessage line and test the customer task.

**Can the agent call my product's API?**
Yes. Connect webhook tools, MCP or OpenAPI sources, or SDK-defined tools. Your backend still determines which customer records the caller may access.

**Can customers pay and approve actions without leaving the thread?**
Eligible iMessage conversations support Keiki Vault Face ID approval cards. With Payments configured and the business enabled, a supported iMessage line and a request eligible for native Apple Pay can deliver a Vault Pay card in the thread. Other requests, including digital goods, use a checkout link. These require the relevant setup, not just an agent prompt.

**What if the agent cannot finish a task?**
Use the team inbox and human takeover. A teammate can reply in the same conversation and return control to the agent with the team's messages preserved.
