State-of-the-art memory, powering Fabric

Fabric now has the best AI memory in the industry

Fabric's memory and search just got a major upgrade.

M-1, our first-generation memory engine, achieved the highest reported score on LongMemEval, the main public benchmark for conversational memory. 96.4% accuracy at top-50 retrieval, outperforming every other reported system.

LongMemEval is a difficult benchmark: roughly 115,000 tokens of conversational history, with relevant information spread sparsely across sessions and buried in noise. It tests six different memory capabilities: recalling user facts, recalling preferences, recalling things the assistant said, synthesizing across sessions, temporal reasoning, and recognizing when information has changed.

M-1 is what powers memory and search inside Fabric. When you search for a note you saved months ago, or Fabric surfaces something relevant you'd forgotten about, or your AI assistant recalls a preference you mentioned in passing, that's M-1 doing the retrieval.

This result validates the approach we've been building towards for years. Memory in Fabric isn't a bolt-on feature. It's core infrastructure, and it's now independently verified as the best performing system of its kind.

We're continuing to develop M-1, and improvements will flow directly into Fabric as they land. Better retrieval, better recall, better search. The product you use every day keeps getting smarter.

Full research paper and methodology here.