Memory
Your team's memory, actually retrievable.
Every index has blind spots, so Stash runs three at once: a curated wiki for structure, vector search for meaning, and grep for exact identifiers. One query hits all three, so your agents retrieve the right thing instead of the closest match.
See the Company Brain →Three indexes, one retrieval.
A question runs against all three and the results merge: keyword precision, semantic recall, and a navigable structure, together.
While you sleep, an agent curates your history into linked pages: a virtual file system your agents navigate directly, the way a teammate opens the right doc.
Every session, page, and table is embedded, so agents find knowledge by meaning, not by remembering the exact filename or wording.
Agents search your Stash like a repo, for the literal identifiers, error strings, and names that embeddings gloss over.
Shared, not per-agent
One memory the whole team and every agent reads from. No per-agent black box, no knowledge trapped in one run's context.
Kept in sync
As your connected sources change, the index updates, so agents reason over the current state of your work, not a stale snapshot.
Cited back to the source
Retrieved knowledge links to where it came from, the page, session, or table, so answers are checkable, not guesses.