MemT 1.0 Keeps What You Teach It Inside the Model, @sudoingX Finds

A model that writes what you tell it into its own memory state, then keeps it across restarts.

MemT 1.0, from the @MemOS_dev team, writes what you tell it into a memory state inside the model for later answers, @sudoingX explains, while the current message still goes in the context window.

He ran it in bf16 on his 128 GB Strix Halo. In the first session he gave it his Rust house rules and one lesson: one error enum called BeastError, no unwrap and the ? operator everywhere, and a missing config must never crash but fall back to defaults with a warning.

He saved its memory to a 42 MB file, killed the process and started a new one from that file. With the same weights, that memory and zero messages in context, he asked for a config loader. It used BeastError on every error path, ? everywhere with no unwrap, and defaults plus a warning for a missing file.

A bigger 288B version is built on DeepSeek V4 Flash, and the weights are open for both, he says. He posted links to the nano and flash weights and to the code and paper.