Concepts
The vocabulary you'll see everywhere in the product, the API, and the rest of these docs.
Mnemos' data model is deliberately small. If a term doesn't appear here, it's either a UI affordance or an extension of one of these primitives.
Organization
An organization is the billing and identity boundary in Mnemos. It owns the SSO connection, SCIM directory, the seat pool, and the audit ledger. Each organization can contain one or more workspaces. Most companies operate a single organization; large enterprises sometimes maintain one per business unit to isolate knowledge graphs while sharing a contract.
Identifier: org_01HZX.... Created at sign-up. Cannot be deleted while active workspaces remain.
Workspace
A workspace is the primary unit of work. It contains its own memory graph, SOP library, members, integrations, and audit scope. Within a workspace, all members share the same knowledge surface — permissions narrow what each member can see and do, but everyone is operating on a single graph.
A workspace usually corresponds to a company. Multi-workspace setups are appropriate when knowledge must be hard-isolated (for example, regulated subsidiaries or M&A scenarios where the buyer and seller can't see each other's data).
Project
A project is a logical grouping inside a workspace: a department (Sales, Finance), a function (Onboarding), or an initiative (Q3 launch). Projects scope membership, retention policies, and access controls without splitting the underlying memory graph. SOPs, sessions, and artifacts each belong to exactly one project.
Member & role
A member is a user with a role inside a workspace. Roles are Owner, Admin, Curator, Contributor, and Viewer. Roles cascade down to projects unless overridden. The full permissions matrix lives in Members & roles.
Memory graph
The memory graph is the structured, queryable model of how your company works. Nodes are entities — people, teams, systems, workflows, decisions, customers — and edges are relationships with type, direction, and provenance. Every edge carries a citation to the source artifact it was extracted from.
The graph is permission-aware: a query that would traverse an edge the requester can't see returns a redacted-but-acknowledged result, never a hallucination around it.
Standard Operating Procedure (SOP)
An SOP is an extracted, versioned, human-reviewable procedure. SOPs have a lifecycle (draft, review, approved, archived), a freshness score, an owner, and zero or more dependencies on graph entities. When an entity changes (a system is deprecated, a team is restructured), every SOP that depends on it is marked stale automatically.
Session
A session is a single AI-led interview, an uploaded recording, or a document ingestion run. Sessions are the canonical input to the extraction pipeline. Each one has a transcript, annotations, extracted candidate SOPs, candidate entities, and an immutable audit record.
Artifact
An artifact is any source-of-truth document attached to a session or referenced by an SOP: a transcript, a slide deck, a recording, a Notion page, a Salesforce record snapshot. Artifacts are addressable by URI and are always cited rather than copied.
Citation & lineage
Every AI answer in Mnemos carries citations to the artifacts and graph nodes it drew from, with character-level offsets where applicable. The full lineage — query, candidate set, permission filter, ranked citations, generated answer — is written to the audit log and is replayable.