openbrainstore.
Inspectable retrieval / one recall, from the inside

Fly through a memory.

Act I / Text

Words strike first.

lanes_run: structured, fts, graph vector_fired: false duration_ms: 9.5 fusion: RRF deep traversal on standby — one flag away
openbrainstore.
Inspectable retrieval

Watch a recall think.

(The cinematic version needs WebGL — here's the same cascade, stage by stage.)

Lanes 01–02 / text

Words strike first.

Structured filters and full-text search light up exact, explainable matches — type, tags, entities, dates, and ranked keyword hits over what your memories literally say. No model in the loop.

Lane 03 / graph

Context assembles itself.

From those hits, MENTIONS and LINKS_TO edges pull in what the matches are connected to — one or two hops, no guessing.

Fusion / RRF

Every lane votes.

score(m) = Σ 1/(60 + rank). A memory two lanes agree on beats any single lane's favorite.

Lane 04 / deep traversal

When words aren't enough, it goes deeper.

A full semantic traversal across everything you've stored — automatic when the fast lanes come up short, instant on deep=true. In embedding space related memories live near one another, so a hit doesn't just find one point — it lights up the whole neighborhood. Always accounted for in the telemetry.

lanes_run: structured, fts, graph vector_fired: false duration_ms: 9.5 fusion: RRF
← back to the architecture