What's the difference between session memory, user memory, and institutional memory?

Three tiers of AI memory

AI memory isn't one thing — it's three tiers with different lifetimes, scopes, and access patterns. Session memory holds the current turn's context: what we just said, the open file, the active task. User memory holds personal preferences and recent projects for a specific human. Institutional memory holds shared truth: org chart, decisions, policies, years of history. Most consumer AI tools collapse all three into a vague "memory" feature. The bRRAIn architecture separates them so each tier gets the storage, refresh cadence, and access control it actually needs.

Session memory is hot and small

Session memory lives in the LLM's context window and the Memory Engine's short-term cache. It holds the last few turns, the active document, the current query intent. Lifetime is minutes to hours. It's expensive per token, so bRRAIn keeps it lean — pointers to the master context rather than inline copies of every document. When the session ends, session memory evaporates; anything worth keeping has already been written to user or institutional tiers by the Consolidator.

User memory is warm and personal

User memory lives in the user's Workspace inside the bRRAIn platform. It holds preferences, recent focus areas, private notes, and ongoing projects. Lifetime is weeks to years. Access is scoped to the user by the Control Plane — teammates cannot read it unless it's explicitly published to a shared workspace. User memory is where you capture "I prefer concise answers" and "I'm currently leading the vault migration" without polluting the institutional graph with personal state.

Institutional memory is cold, shared, and canonical

Institutional memory lives in the bRRAIn Vault as the company's canonical graph: every Person, Organization, Place, Event, decision, and policy. Lifetime is indefinite. Access is role-gated across the 7-tier hierarchy. It updates through the Consolidator, not directly, so writes are validated and merged. At session boot, the Memory Engine assembles a consolidated master context from all three tiers — session context (fresh), user context (warm), institutional context (cold) — and hands it to the LLM as one payload. That's how a stateless model answers like it remembers.

Relevant bRRAIn products and services

• Memory Engine — assembles session, user, and institutional tiers into one context.
• Workspaces — storage for warm user memory, scoped per person.
• bRRAIn Vault — cold institutional memory with envelope encryption.
• Consolidator — promotes session writes up the tiers with conflict resolution.
• Control Plane — role-scoped access across all three tiers.