Back to directory

Cardputer AI

A fully offline chatbot that runs entirely on an ESP32-S3 Cardputer microcontroller, with an 8-million-parameter model quantized to 4 bits and baked into the firmware.

Cardputer AI

Cardputer AI is a chatbot that runs entirely on a Cardputer ADV, a pocket-sized microcontroller device with 512 KB of RAM and no internet connection. Its language model has 8 million parameters, quantized to 4 bits, stored as a 5.7 MB file baked directly into the firmware, and it generates about 5 tokens per second.

Unlike earlier tiny on-microcontroller language models, which could only continue a story once given an opening line, this one holds a back-and-forth conversation: it stays on topic, asks follow-up questions, and says it does not know an answer rather than inventing one. It also supports a separate story-writing mode that continues a prompt as a short narrative.

The firmware is downloadable and requires no cloud connection, no phone and no companion app.

More in Inference

See all 770
  • OpenDecision

    deepanwadhwa.github.io

    An open-source Python package that answers typed questions about application state or documents using a local natural-language-inference model, returning structured choice, score, yes/no or evidence-relation values instead of free text.

  • Lifeboat

    iterate.ai

    A local inference server that puts an OpenAI- and Anthropic-compatible API in front of open-weight language models running on hardware you control, from a laptop to a GPU cluster.

  • Claude S40

    github.com

    An unofficial Claude chat client for 2007-era Nokia Series 40 phones, built as a Java ME app plus a small Go relay server, with web search, reading mode and saved chats.