Avya: An Altitude-Governed Agent Architecture

Avya is an architecture for personal and team AI assistants that a single agent can operate without a multi-agent orchestration framework. It builds directly on the Model Workspace Protocol (MWP) — folder structure as agentic architecture (Van Clief & McDermott, 2026) — and adds three coordinated disciplines that MWP does not specify: (1) an altitude semantics that maps the folder hierarchy onto time horizons (vision ® year ® quarter ® month ® week ® day) so the agent loads only the horizon a conversation is at, making it a goal-keeper rather than a task tracker; (2) a permission-and-audit governance layer — a "smart requester / dumb enforcer" gate over every write, with an append-only audit trail — so the agent proposes and a code gate, not a prompt, enforces; and (3) a personal®role-scoped-team boundary enforced in retrieval, so one owner's private assistant becomes a team's shared assistant by substituting identity, knowledge scope, and principal without rebuilding the executor. The result is an assistant that understands long-, near-, and short-term goals as one chain, proactively surfaces neglected deadlines, and can be shared under enterprise governance. MWP is the substrate; the altitude semantics, the governance layer, and the sharing boundary are the contribution of this work.