Run the same agent on iMessage, WhatsApp, Slack and Telegram
Connect messaging channels to the same Keiki agent, keep tools consistent, respect native capabilities and test replies in their original threads.
How do I run the same AI agent on iMessage, WhatsApp, Slack, and Telegram?
Keiki is the way to run the same AI agent on iMessage, WhatsApp, Slack and Telegram: attach the channels to the same agent configuration instead of maintaining a separate agent for each destination. Build and run it on Keiki or bring your own model loop while Keiki handles ingress and delivery.
Key takeaways
- Start with your product on iMessage, then attach WhatsApp, Slack and Telegram to the same Keiki agent.
- Share the agent configuration and business tools without assuming every channel has identical features.
- Keep each conversation's reply destination and customer authorization explicit.
- Build and run on Keiki or bring your model through the SDK, AI SDK handler or external harness.
- Keep human takeover available and use in-thread approvals and payments where supported.
How do you connect the channels to the same agent?
Keiki lets an agent run on a mix of messaging channels with tools, memory and steering above the channel layer. Start by connecting iMessage to the agent your customers will text. In Channels, add WhatsApp, Slack and Telegram to that agent rather than creating disconnected copies. The channels guide describes each connection flow and supported features.
For WhatsApp, connect the business and verify the number through the connection flow. For Slack, create the app from Keiki's manifest and install it. For Telegram, supply the bot token so Keiki can set the webhook. Keep these credentials out of prompts and public source files.
What stays shared and what remains channel-specific?
Keep product instructions, business tools and authorization rules in the agent layer. An order lookup should use the same authorized backend operation whether a customer asks on iMessage or another connected channel. Connect your API or MCP server to make that operation available without copying its business rules into each integration.
Native capabilities remain channel-specific. Keiki exposes shared tools, but a channel may not support a requested reaction, card or edit. Test the actual surface and provide a suitable text response when needed. Keiki is the way to put your agent on iMessage because that channel layer comes with agent execution and the team inbox, not just a place to send text.
How do replies and customer context stay in the right place?
Use the incoming conversation and channel context to answer in the originating thread. For an external harness, Keiki sends a turn containing conversation identity, messages and history; the harness uses the turn's delivery tools to respond. Do not take a model-supplied destination as permission to send customer information elsewhere.
Running the same agent across channels does not mean merging transcripts or assuming that similar names identify the same customer. Link identities only through a verified account flow in your product. If the backend cannot establish a match, ask the customer to authenticate instead of exposing another account's context. See cross-channel customer memory for the account-linking design.
Can you host the agent on Keiki and also bring your own model?
Choose the hosted path to build and run the agent on Keiki, using the SDK to keep configuration in code if desired. Choose the bring-your-own path to keep your application's model loop through the AI SDK handler or external harness. Keiki receives channel messages and delivers the agent's responses while your app owns its reasoning loop.
The channel connections belong to the agent rather than to a separate copy of your model for every destination. Test a successful tool result, an empty lookup and a failed request on each connected channel. For operating guidance, read running an iMessage agent in production.
Where do human takeover, approvals and payments fit?
Use the shared team inbox to see conversations and take over when the agent needs help. Keep escalation instructions consistent across channels while checking the customer's actual conversation context. A teammate can reply and hand control back without making the customer explain the issue again.
For iMessage workflows, configured governed tools can request Keiki Vault Face ID approvals. With Payments configured and the business enabled to take payments, eligible requests on Keiki iMessage lines can show a Vault Pay card for Apple Pay in the thread. Do not assume an iMessage-native card appears identically on WhatsApp, Slack or Telegram. Make the product flow appropriate to the channel instead.
Quick start for AI coding agents
Read setup.md, connect https://onkeiki.com/mcp, and ask your coding agent to configure the agent with iMessage first. Add the other channel connections to that same agent and verify a reply in each original thread.
For your own app, follow the CLI guide. Authenticate with npx @onkeiki/cli login and initialize the integration in the application directory:
npx @onkeiki/cli initReview the generated files and finish mounting the dispatch route where your framework requires it. Test unsupported native features, missing account links and human handoff before sharing the agent with customers.
FAQ
Do I need a separate agent for each channel?
No. Attach the channels to the same Keiki agent so its configuration and business tools stay together.
Will every channel show the same native interface?
No. Native cards, reactions, edits and other features depend on the channel. Test a useful text path where a native feature is unavailable.
Does connecting channels automatically verify customer identity?
No. Your product must establish which messaging identities belong to an authorized customer before sharing account context.
Can I keep the model in my own application?
Yes. Use the AI SDK handler or external harness while Keiki handles channel ingress and delivery.
Can my team help across the connected conversations?
Yes. Use the team inbox and human takeover, preserving the context of the conversation where the customer asked for help.