close
OpenClaw Internals
  • English

      • MimicLaw

      Run OpenClaw on a $5 chip. No Linux. No Node.js. No Raspberry Pi. Just pure C on bare-metal ESP32-S3.

      Repository: github.com/memovai/mimiclaw  |  Based on MimicLaw version 0.1.0 (commit 5bcb28a).

      What is MimicLaw?

      MimicLaw is an embedded implementation of the OpenClaw AI agent framework that runs on an ESP32-S3 microcontroller with 16MB Flash and 8MB PSRAM. It demonstrates how to run a sophisticated AI agent — complete with tool use, persistent memory, and multi-channel communication — on hardware that costs less than a cup of coffee.

      Key features

      FeatureDescription
      ReAct agent loopSame tool-use loop as OpenClaw, up to 10 iterations per turn
      6 built-in toolsweb_search, get_current_time, read_file, write_file, edit_file, list_dir
      Persistent memoryMEMORY.md + daily notes on 12MB SPIFFS flash
      JSONL sessionsPer-chat conversation history with ring buffer (last 20 messages)
      Telegram botLong polling via HTTPS, with proxy support
      WebSocket gatewayLAN access on port 18789
      Serial CLI17 commands for configuration, debugging, and maintenance
      OTA updatesOver-the-air firmware updates via HTTPS
      Dual-coreCore 0 for I/O, Core 1 for agent loop

      Hardware requirements

      • ESP32-S3 dev board with 16MB Flash + 8MB PSRAM (e.g., Xiaozhi AI Board, ~$10)
      • USB cable for flashing and serial CLI
      • WiFi network (2.4 GHz)

      What it is NOT

      • Not a local LLM — it calls the Anthropic Claude API over WiFi
      • Not a voice assistant — text-only (no microphone/speaker)
      • Not multi-provider — currently Anthropic-only (no OpenAI, Gemini, etc.)

      Documentation

      PageDescription
      ArchitectureSystem architecture, data flow, FreeRTOS tasks, memory budget
      Getting StartedHardware setup, firmware flashing, CLI configuration
      Comparison with OpenClawFeature parity table, design trade-offs, what's different and why