MimiClaw - An open-source, ultra-lightweight AI assistant that requires no advanced runtime environment.

What is MimiClaw?

MimiClaw isBased on ESP32-S3 chipofUltra-lightweight AI assistantWritten in pure C, it can run directly on bare metal.It requires no Linux, Node.js, or operating system. MimiClaw provides 24/7 online access, delivering Claude/GPT intelligent services via Telegram or WebSocket. MimiClaw features a persistent memory system (SOUL.md/USER.md/MEMORY.md), tool invocation support for search/scheduled tasks, and hardware control capabilities. All data is stored locally with privacy as a priority, representing the ultimate engineering practice of edge AI and agent hardwareization.

MimiClaw's main functions

• Intelligent dialogueReal-time chat with Claude/GPT models via Telegram Bot or WebSocket, supporting dual-Provider runtime switching.
• Local memoryMimiClaw persistently stores AI personalities, user information, and long-term memories in Markdown file format, ensuring that they are not lost upon power failure and restart.
• Autonomous schedulingIt features a built-in Cron scheduler and heartbeat mechanism, enabling AI to create scheduled tasks and proactively check and execute pending items.
• Tool callSupports tools such as network search, time acquisition, and GPIO hardware control to implement Agent loop in ReAct mode.
• Wireless Operations and MaintenanceSupports OTA firmware updates and HTTP proxy, enabling deployment and network adaptation without a physical connection.

MimiClaw's technical principles

• Bare metal architectureDeveloped in pure C language based on the ESP-IDF framework, it runs directly on ESP32-S3 hardware without requiring Linux or Node.js runtime. By reducing dependencies, the code size is kept within 16MB of Flash memory.
• Dual-core task allocationUsing the dual-core Xtensa LX7 processor of ESP32-S3, the network I/O protocol stack is bound to CPU0, while agent inference and tool execution are placed on CPU1 to achieve concurrent processing and avoid blocking.
• Streaming memory managementTo address the 512KB SRAM limitation, HTTP streaming and chunked JSON parsing are employed to avoid memory overflow caused by loading large responses at once. Additionally, temporary data is stored using PSRAM.
• ReAct state machineImplement a ReAct (reasoning-action-observation) loop state machine in pure C, maintain multi-turn dialogue context, parse the JSON tool call instructions returned by the LLM, and drive the execution of local tools.
• SPIFFS persistence: Establish a SPIFFS file system on Flash to store memories, configurations, and task queues in plain text format, achieving zero-dependency data persistence and human editability.

MimiClaw project address

• Project official websitehttps://mimiclaw.io/
• GitHub repositoryhttps://github.com/memovai/mimiclaw

Application Scenarios of MimiClaw

• Personal assistantAs a pocket-sized AI companion, it can be plugged into a power bank or car USB port to be online 24/7. You can check your schedule, record inspiration or get advice anytime via Telegram without relying on your phone's computing power or cloud subscriptions.
• Smart Home HubAs a low-power home gateway, it connects to GPIO devices such as temperature and humidity sensors and relays, controls home appliances through natural language commands, and locally memorizes user habits to achieve personalized automation.
• Industrial edge nodesDeployed in scenarios such as factories or farmlands, it collects equipment data on a regular basis and autonomously judges anomalies. It actively reports through a heartbeat mechanism. Its 0.5W power consumption is suitable for unattended environments powered by solar energy.
• Educational Development PlatformThe open-source bare-metal code provides embedded learners with sample Agent engineering practices, and the low-cost hardware lowers the barrier to AIoT development, making it suitable for university IoT and edge computing courses.