Hardware messaging assistants
Telegram messages reach an assistant on a compact board, which retains conversation records and coordinates tasks while an online model generates responses.
Community Meetup · Singapore · 14 April 2026
Running the assistant on a small board
MimiClaw runs in C on ESP32-S3 without Linux or Node.js. The demonstrated ESP32-S3-DevKitC-1 N32R16V had 32 MB flash and 16 MB PSRAM. Model inference used a remote API.
Sessions, buffers and message queues
Local session files stored conversations. PSRAM held prompt, history and tool output buffers, while queues handled incoming Telegram messages and outgoing responses.
Connecting and diagnosing the device
ESP-IDF v6.0 supplied the firmware toolchain. WiFi connected Telegram and the remote model API. Setup included rebuilding and flashing firmware, then monitoring the serial port.
Troubleshooting covered USB versus UART, SSL certificates, unstable connections and model latency.