Hindsight auto-organizes your activity into Mental Models, Observations, World Facts, and Experience Facts, and retrieves with a multi-strategy approach combining semantic, BM25, graph, and temporal signals. Brain OS captures structured operational state (decisions, plans, blockers, focus) through typed MCP tools. Different categories of AI memory.
Hindsight (by Vectorize, Inc) sits in the transcript-and-activity-capture category, but with more sophistication than most. It auto-organizes what it sees into four kinds of memory (Mental Models, Observations, World Facts, Experience Facts) and retrieves them using TEMPR, a multi-strategy approach that combines semantic search, BM25 lexical search, graph traversal, and temporal signals.
It is a commercial product designed to give agents broad, cognitively-organized memory across many domains. The mental model: a structured-ish memory layer the system curates for you, with rich retrieval over what it has captured.
Brain OS does not auto-organize anything. Your agent writes operational state intentionally: decisions with reasons and rejected alternatives, plans with phases, blockers, focus, momentum, and recurring patterns. The categories are not cognitive science abstractions, they are the operating primitives of doing work: what did we decide, what are we doing, what's in the way, what matters today.
The difference: Hindsight answers "what have we observed about this domain?" Brain OS answers "what is true about this project right now?"
.brain/ shared by Claude Code, Cursor, Zed, GitHub Copilot, any MCP clientdecision_check returns conflict/caution/clear with prior reasoningyou want your agent to build broad, organized memory across many domains and activities, and you want a commercial product handling the curation and retrieval pipeline.
you want your agent to operate from your project's current state: decisions, plans, blockers, focus, captured intentionally as you work and exposed through typed tools. Local-first, MIT, no curation pipeline.
The two are complementary, not competing. They sit on different axes, cognitive knowledge vs operational truth.