lossless-memory is the counter-position: never summarize, keep every line, timestamp everything.

62 HN points on September 21, 96 stars, created September 4 and last pushed September 5, so treat the traction as one data point rather than adoption. It matters as an argument in a category saturating fast, where deja-vu, ai-memory, supermemory, mempalace and claude-mem all compete on how aggressively to compress session history. Read it next to the VibeMemBench result above, where eleven of twelve retrieval layers lost to no memory at all. (GitHub)