An AI assistant that fits in your pocket - literally

The team behind MimiClaw set out to prove that a personal AI assistant doesn't need a server, a subscription, or even a full operating system. MimiClaw runs entirely on an ESP32-S3 microcontroller - a board you can buy for $5, power from a USB cable, and hold between two fingers.

The assistant handles real tasks: it schedules reminders, answers questions, manages to-do lists, and remembers what you told it weeks ago - even after a reboot. It connects to Claude or ChatGPT over WiFi, but all the orchestration - memory, task scheduling, tool use - runs locally on the chip. No cloud dependency for the core logic. No monthly infrastructure cost. Just 0.5 watts, always on.

The project resonated. MimiClaw crossed 4.4k GitHub stars, drew coverage from Hackster, CNX Software, and TechStrong AI, and built an active community of developers and makers. But it had one limitation that kept holding it back: the only way to talk to your assistant was through Telegram.

For most users - especially in the US - Telegram is an extra app. It's not where they text friends, coordinate plans, or check notifications. MimiClaw's whole promise was an AI companion you could reach anytime. But if reaching it required switching to a separate app, the "anytime" part didn't quite land.

From a $5 chip to iMessage - without changing a line of core code

MimiClaw architecture overview

Photon gave MimiClaw what it couldn't build alone: a direct line into iMessage.

Building native iMessage support from scratch means dealing with macOS system permissions, Apple's messaging protocols, message delivery infrastructure, and thread management - months of work that has nothing to do with building a better AI assistant. Photon handles all of it.

With Photon's iMessage integration, MimiClaw added iMessage as a first-class channel alongside Telegram. The assistant receives messages through Photon's API, processes them through the same AI pipeline it already uses, and sends responses back through iMessage - with full thread continuity, native formatting, and reliable delivery.

From MimiClaw's side, the integration was clean. Here's the entire configuration:

#define MIMI_SECRET_IMSG_SERVER_URL "<https://xxxxx.imsgd.photon.codes>"
#define MIMI_SECRET_IMSG_API_KEY    "your-api-key"
#define MIMI_SECRET_IMSG_PROXY_URL  "<https://imessage-swagger.photon.codes>"

#define MIMI_SECRET_IMSG_SERVER_URL "<https://xxxxx.imsgd.photon.codes>"
#define MIMI_SECRET_IMSG_API_KEY    "your-api-key"
#define MIMI_SECRET_IMSG_PROXY_URL  "<https://imessage-swagger.photon.codes>"

Three lines. That's all MimiClaw needed to tell Photon where to connect. From there, Photon handles auth, message polling, delivery, and thread management automatically.

Once connected, sending a message through iMessage is a single function call:

imessage_send_message(chat_id, text

imessage_send_message(chat_id, text

And routing inbound messages into MimiClaw's existing agent loop took one new case in the message dispatcher:

} else if (strcmp(msg.channel, "imessage") == 0) {
    imessage_send_message(msg.chat_id, msg.content

} else if (strcmp(msg.channel, "imessage") == 0) {
    imessage_send_message(msg.chat_id, msg.content

No changes to the AI logic, memory system, or task scheduler. The assistant doesn't know or care whether you texted from iMessage or Telegram - it just responds. That's the point: Photon works with any system, so teams can focus on their product instead of messaging infrastructure.

Why iMessage makes hardware AI real

Telegram was the right starting point - open API, easy to integrate, works everywhere. But for a personal AI assistant that's supposed to feel like it's yours, the messaging app matters. iMessage is private, familiar, and already part of daily life. People don't "open iMessage" - they just text. When an AI assistant lives there, it doesn't feel like a product you have to learn. It feels like someone you can reach. That shift - from "I'll check my AI app" to "I'll just text it" - is the difference between a novelty and a habit. For MimiClaw, moving to iMessage turned a maker project into something people actually use every day.

Early traction and what's next

MimiClaw is growing fast. In less than one month, 4.4k GitHub stars, 611 forks, and a community of developers building everything from home automation agents to personal productivity assistants on the platform.

iMessage support opens MimiClaw to a much larger audience - the millions of iPhone users who want a personal AI assistant but don't want another app on their home screen. And because Photon supports additional channels beyond iMessage, MimiClaw has a clear path to WhatsApp, SMS, and more - without rebuilding the messaging layer each time.

With Photon handling messaging infrastructure, the MimiClaw team can focus entirely on making their assistant smarter, more capable, and more personal. The chip stays tiny. The memory gets deeper. And iMessage makes it all feel as natural as texting a friend.