Frequently Asked Questions

OpenClaw is an open-source AI agent platform that runs on ESP32 microcontrollers. Unlike a simple chatbot, it's an agent that can reason about which tools to use and interact with the physical world — reading sensors, controlling devices, and communicating through platforms like Telegram and Discord. The cheapest setup costs about $5 in parts.

MimiClaw is the minimal variant designed for the ESP32-C3 ($2 chip, 400KB RAM). ESP-Claw is the full-featured variant for the ESP32-S3 with 8MB PSRAM, supporting voice I/O and local ML inference. Both run the same AI agent core with 40+ tools.

Yes, all software and hardware designs are open source under MIT (software) and CERN-OHL-P (hardware) licenses. You only pay for the physical components ($5-25) and AI API usage (~$0.50/month for typical home use with Claude Haiku).

Not for basic setup. You can flash firmware from your browser, configure via a web interface, and customize your AI's personality by editing a Markdown file (SOUL.md). Programming in C or MicroPython is only needed if you want to create custom tools.

For beginners on a budget: ESP32-C3 SuperMini (~$2). For the full experience with voice: ESP32-S3 DevKitC N16R8 (~$8). Both are available on AliExpress and Amazon. See our Products page for a detailed comparison.

Currently, the original ESP32 (ESP32-D0WDQ6) is not supported. ESP-Claw is optimized for the newer C3, S3, and C6 variants which have better power efficiency and security features.

About 0.5 watts when active (ESP32-C3). That's less than an LED nightlight. Running 24/7 for a year costs approximately $0.50 in electricity. With deep sleep, a 1000mAh battery can last weeks to months.

Yes. ESP-Claw supports any sensor that connects via GPIO, I2C, SPI, or analog input. Common additions include temperature sensors (DHT22, BME280), motion detectors (PIR), displays (OLED), and servo motors. Check our Compatible Hardware list for tested devices.

Yes, using an IR LED (~$0.10). ESP-Claw can learn your AC remote's IR codes and then control the AC via natural language commands. Say 'set AC to 24 degrees' and it sends the appropriate IR signal.

Claude (Anthropic), OpenAI (GPT-4, GPT-3.5), Ollama (local models), and any OpenAI-compatible API endpoint. Claude Haiku is recommended for the best cost/performance ratio at about $0.25/million input tokens.

Your SOUL.md, MEMORY.md, and device configuration stay on the ESP32's local flash — they never leave your device. When you send a message, only that conversation is sent to your chosen AI provider (Claude, OpenAI, etc.) for processing. You can use a local Ollama server for fully offline operation on ESP32-S3.

Typical home use with Claude Haiku costs about $0.30-0.50 per month (roughly 10-20 conversations per day). A single conversation with 2-3 tool calls uses about 500-1000 tokens, costing $0.0003-0.001. You can set spending limits in your API provider's dashboard.

Partially. The ESP32-S3 can run small TensorFlow Lite models locally for wake word detection and simple intent classification. For full conversational AI, you need either a cloud API or a local Ollama server on your network. The ESP32-C3 always requires a cloud API for AI inference.

The easiest way is our browser-based flasher at the Install page — just connect your board via USB, click 'Flash' in Chrome or Edge, and you're done in 30 seconds. No software installation required. You can also use esptool.py or build from source with ESP-IDF.

Most common cause: a charge-only USB cable. Try a different cable that supports data transfer. You may also need to install USB drivers (CH340 for ESP32-C3 SuperMini, CP2102 for some S3 boards). On macOS, check System Settings > Privacy > USB.

Wait 10 seconds after the flash completes. The LED should blink blue when the hotspot is active. If nothing happens, try pressing the Reset button. If the LED blinks red, the flash may have failed — try flashing again.

Check: 1) Is the ESP32 connected to Wi-Fi? (LED should be solid, not blinking). 2) Is the bot token correct? 3) Is your AI API key valid and has credit? 4) Check the serial monitor for error messages by connecting via USB.

Yes, ESP-Claw supports OTA (over-the-air) updates via Wi-Fi. Tell your bot 'check for updates' or configure automatic update checks. See our OTA Updates guide for details.

SOUL.md is a plain Markdown file that defines your AI agent's personality, behavior rules, and knowledge. It's the 'instruction manual' that the AI reads before every conversation. You can edit it through the web interface or upload a new file. Changes take effect immediately without reflashing.

Yes. ESP-Claw uses MEMORY.md to store information across conversations and reboots. The AI can save user preferences, learned routines, and contextual notes. You can also define what it should and shouldn't remember in SOUL.md.

Yes. Tools are C functions that you register with the agent during initialization. Each tool has a name, description (for the AI to understand when to use it), and a handler function. See the API Reference for the claw_tool_register() function.

Yes. Each ESP32 board runs one agent, but you can deploy multiple boards with different SOUL.md personalities. They can communicate with each other via MQTT, creating a multi-agent system — for example, a climate agent in each room reporting to a central hub agent.