# memovai/mimiclaw

MimiClaw: Harness on a $5 chip. No OS(Linux). No Node.js. No Mac mini. No Raspberry Pi. No VPS. Hardware agents OS.

Repository: https://github.com/memovai/mimiclaw
Canonical: https://ross.abutalabs.com/products/mimiclaw
Homepage: https://mimiclaw.io
Language: C
License: MIT
License Family: permissive
Topics: clawdbot, memory, openclaw, ai, assistant, edge-ai-agents
Last push: 2026-08-21T14:17:29+00:00

## Health v2 (maintenance only)
Score: 69/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 98, release rhythm 63, longevity 15
- inputs: {"age_days": 210, "days_push": 12, "days_rel": 169, "gap_med": 35, "n_releases_24m": 2}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 5731, forks 891 (observed 2026-08-28T04:09:28.318383+00:00)

## What it is
MimiClaw is a bare-metal AI agent framework written in pure C that runs an OpenClaw-style personal AI assistant on a $5 ESP32-S3 microcontroller with no OS or Node.js. It connects to Telegram over WiFi, runs a ReAct agent loop with Anthropic or OpenAI LLMs, calls tools, and persists memory in flash across reboots at 0.5W.

## Use cases
- run an AI assistant on a microcontroller without Linux or a server
- build a Telegram-controlled AI agent on cheap hardware
- deploy an always-on edge AI agent with persistent memory
- control GPIO sensors and actuators from an LLM agent
- host a personal AI assistant on a $5 chip instead of a VPS
- flash an OpenClaw-compatible agent onto ESP32-S3

## When to choose
- you want a tiny, low-power, always-on AI agent on dedicated hardware
- you prefer bare-metal C with no OS, runtime, or cloud server dependency
- you need local persistent memory that survives reboots and stays on-device
- you want hardware interaction (GPIO, sensors) driven by an LLM agent

## When to avoid
- you need heavy local LLM inference — it relies on cloud Anthropic/OpenAI APIs
- you need a full OS, rich libraries, or Node.js ecosystem tooling
- you don't want to buy an ESP32-S3 board or use ESP-IDF build tooling
- you need high-throughput or multi-user server-side agent workloads

## Facets
- artifact type: framework
- maturity: active
- function: agent-framework, llm-inference, chatbot, embedded, websocket, http-client
- domain: artificial-intelligence, large-language-models, embedded-systems, iot, chatbots, hardware
- platform: embedded, iot, cross-platform
- tags: esp32-s3, bare-metal, edge-ai, telegram-bot, openclaw, esp-idf, pure-c, microcontroller, persistent-memory, hardware-agent, memory, ai-agents, telegram

## Member repositories
- memovai/mimiclaw (main) score 69

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:28.318383+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:53:23.611314+00:00, confidence not recorded.
  - readme: https://github.com/memovai/mimiclaw (fetched 2026-08-28T04:09:28.318383+00:00, sha f24c7b8f526a)
  - homepage: https://mimiclaw.io (fetched 2026-08-29T08:48:47.903251+00:00, sha 076591e48f2d)
- Data as of 2026-08-30T08:39:29.467469+00:00.