Show HN: A memory database that forgets, consolidates, and detects contradiction

A new “memory” database called YantrikDB just hit the stage promising to forget old stuff, merge duplicates, and flag contradictions so your AI agent doesn’t drown in its own notes. The pitch: real memory isn’t a junk drawer; it should decay over time, compress 20 near-duplicate meeting notes into a handful, and scream when “CEO is Alice” collides with “CEO is Bob.” There’s a server, a plug-in mode for AI coding tools, and an embeddable library—and the dev flexes chaos tests, encryption, and maturity notes.

But the comments? Pure split-screen. One camp is all-in on the half-life idea and wants the juicy details: “What’s the consolidation loop—random sampling and an AI merge?” Another camp is pounding the table for proof: benchmarks, real-world wins, and fewer diagrams. The sharpest jab called it a “slopcoded brainwave,” while a more measured skeptic asked, “Did you check if this leads to any actual benefits?” The author fires back with lived pain: memory in a popular vector store turned “to garbage at ~5k” notes, causing outdated facts and contradictions.

Meanwhile, drive-by comedians are loving the premise: “Finally, a database that forgets my meetings,” and “Can it forget my sprint tickets too?” Under the jokes lies a bigger fight: Do we really need a complex ‘thinking’ memory layer, or should large language models handle it at prompt time? Whether it’s the next brain for agents—or just a neat science project—the crowd wants receipts, not vibes.