memovai/mimiclaw

★ 5,412CAudience · developerComplexity · 5/5Setup · hard

Mindmap

mindmap
  root((repo))
    What it does
      AI on a microchip
      Telegram interface
      Agent loop with tools
    Hardware
      ESP32-S3 board
      USB power only
      On-chip flash memory
    AI Providers
      Anthropic Claude
      OpenAI GPT
      Switch at runtime
    Configuration
      Build-time header
      Serial CLI overrides
    Use Cases
      Edge AI experiments
      Low-power assistant
      Persistent memory

mindmap root((repo)) What it does AI on a microchip Telegram interface Agent loop with tools Hardware ESP32-S3 board USB power only On-chip flash memory AI Providers Anthropic Claude OpenAI GPT Switch at runtime Configuration Build-time header Serial CLI overrides Use Cases Edge AI experiments Low-power assistant Persistent memory

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Build a persistent AI assistant that runs on cheap bare-metal hardware without a cloud server or Raspberry Pi.

USE CASE 2

Experiment with edge AI by running an LLM-backed agent loop on a microcontroller.

USE CASE 3

Create a low-power AI device that remembers conversations across reboots using on-chip flash storage.

USE CASE 4

Switch between Claude and GPT at runtime via a serial command without recompiling firmware.

Tech stack

CESP32-S3Anthropic APIOpenAI APITelegram Bot API

Getting it running

Difficulty · hard Time to first run · 1day+

Requires an ESP32-S3 board, C toolchain, and API keys for Claude or GPT, no operating system or Linux runtime needed.

License not specified in the explanation, check the repository for terms.

In plain English

MimiClaw turns a tiny, inexpensive microcontroller board, the ESP32-S3, which costs around $5 to $10, into a personal AI assistant that runs entirely without an operating system like Linux or a runtime like Node.js. The whole thing is written in C, keeping it lean enough to fit on a chip the size of a thumb. The way it works: you send a message through Telegram, the board picks it up over WiFi, runs it through an agent loop where a large language model thinks and calls tools, then sends the reply back to you on Telegram. It supports both Anthropic (Claude) and OpenAI (GPT) as the AI provider, switchable at runtime without recompiling. All conversation history and memory are stored locally on the chip's built-in flash storage, so it remembers things across reboots. The device draws only 0.5 W and can run continuously from USB power. Configuration uses a two-layer system: build-time defaults are set in a header file, and runtime overrides, WiFi credentials, API keys, AI provider, model choice, and optional search integrations, are entered through a serial command-line interface and saved to NVS flash. This means you can change settings by plugging in a USB cable anywhere, without recompiling. You would reach for MimiClaw if you want a persistent AI assistant running on cheap, low-power hardware without paying for a cloud server or needing a Raspberry Pi or Mac mini. It targets hardware tinkerers and edge AI experimentation. The full README is longer than what was provided.

Copy-paste prompts

Prompt 1

I am setting up mimiclaw on an ESP32-S3. Walk me through entering the WiFi credentials and Anthropic API key via the serial interface.

Prompt 2

How do I add a new tool to the agent loop in mimiclaw's C codebase so the AI can call it during a conversation?

Prompt 3

I want to switch mimiclaw from Anthropic Claude to OpenAI GPT at runtime. What serial commands do I need to send?

Prompt 4

Help me configure the two-layer settings system in mimiclaw, what goes in the header file at build time versus what I set via serial at runtime?

Prompt 5

What is the minimum hardware needed to run mimiclaw, and how do I flash the firmware to an ESP32-S3 board for the first time?

Open on GitHub → Explain another repo

← memovai on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.