daosdaosdaosdaosdaosdaosdaos

DAOS asks a local AI model once, tidies and signs the answer, and gives it back byte for byte when the exact same request comes again. A logged run can be replayed later, a stored answer can be checked against a stronger model, and a single wrong one can be withdrawn or corrected.

We keep it as a research bench rather than a product. On one machine, turning the model’s randomness off was already enough to get one answer, and voting across several copies of a model made answers consistent but not more accurate. The research note covers what we measured, what did not pay off, and when you do not need any of this.

It is a local, single-user Python library that drives models through LM Studio. The source repository is currently private.