RelayCommon questions
Does Relay read everything I type into Claude?
No. Retrieval reads from the shared memory to bring context in; it never ships your conversation out. Writing only happens when you intend to save something.
Is recall really automatic?
On Claude Code, yes — an installed hook searches team memory on its own (at session start and on topic shifts, throttled). On Claude.ai there are no hooks, so recall is driven by Claude following the Relay skill; if it doesn’t surface on its own, just ask it to check team memory. See Retrieval.
When does it save? Does it save every message?
No — saving is intentional, never per-message, and identical on both surfaces. Claude proposes a save and confirms, or you say “share this with the team” to save in one shot. See Saving.
What is mem0’s role here?
mem0 is the underlying vector store that holds the embedded facts and powers semantic search. Relay wraps it with tenant isolation, role scoping, expiry, auditing, and the team-shared model — the parts that make it safe for a team.
Can a teammate see memories meant for another role?
No. Audience scoping is enforced server-side. Only admins see everything; dev and commercial each see the shared (all) memories plus their own scope.