AI assistants are all the rage these days, but they are typically unaware of automation settings and can have difficulty processing tasks asynchronously. Enter zclaw. You have the option to use your personal assistant on an ESP32 that supports Anthropic, OpenAI, or OpenRouter. It totals 888 KB and does not host LLM, but adds important functionality for monitoring and controlling devices connected to the ESP32.
Communicate with your assistant via telegram. You can say something like, “Don’t forget the garage sensor is on GPIO 4.” Then it might say, “Check the garage sensor in 20 minutes, and if the sensor is high, set GPIO 5 lower.” It has an RTOS for scheduling tasks and is aware of time zones and common time periods. Memory persists across reboots and allows you to choose different personas.
Some of the use cases mentioned in the manual show how beneficial it is to have something that can accurately schedule, control, or monitor devices. Ideas like suggesting a lab setup or setting a watering schedule for your plants are difficult to accomplish with just a standard chatbot.
AI can also be self-reflective. For example, you can create some tasks based on a schedule and ask the device to “Show schedule”. You can also create up to 8 tools with names, descriptions, and actions. This allows you to write something like “power_down_bench” and tell zclaw to run it on demand or on a schedule. All in all, an interesting and well-documented setup.
We’ve seen many projects like this, and each one has its own charm. And its own personality.
