News Summary

• Compresses high-dimensional vectors to 2-4 bits per coordinate with near-optimal distortion.
• Data-oblivious (no training), zero indexing time.
• Each vector is encoded independently in ~4ms at d=1536.
• A 1536-dim vector goes from 6,144 bytes (FP32) to 384 bytes (2-bit) That's 16x compression.
• Instead of decompressing every database vector, we rotate the query once into the same domain and score directly against the codebook values.