Hindsight

OrganizationVectorize

Founded2025-10 (repo)

Funding$3.6M seed (True Ventures), 2024-10

CodeMIT · 18.8k★

Four-network memory substrate (world/experience/opinions/entities) with retain/recall/reflect.

Every number on file

claimed	benchmark	config	runner	status
91.4 → 94.6	LongMemEval	reader/judge undisclosed on dashboard	self + self-run leaderboard	SELF-REPORTED

The 'Agent Memory Benchmark' it tops is operated by Vectorize itself — '#1 on every dataset' is a self-run aggregator.

Rows are not comparable — configs differ. A row graduates to a verdict only through a registered re-run. The protocol →