Core boundary · Voice v1.0

LINA

LINA is the interaction and voice runtime layered on AIJIM Core: conversational, operational, and local where it should be, but no longer a parallel audit stack.

What LINA Is Now

LINA remains the voice- and interaction-facing personality of the AIJIM system, but the architectural framing is now sharper than it used to be. LINA is not a second Core and not a separate owner of normative audit truth. LINA is the runtime that turns Core-backed state into interaction, explanation, navigation, and voice.

That means the system now has a clearer split of concerns: Core owns audit-grade persistence and invariant-bearing writes; LINA owns the assistant-local experience around those facts.

Boundary summary

The canonical owner split is documented indocumentation/runbooks/LINA_CORE_BOUNDARY_CONTRACT.mdand the human-readable overview lives at /architecture/lina-core-boundary.

What Stays Local to LINA

LINA-local surface	Why it stays local
Session events	Assistant-local sequencing, wake-state, and interaction flow.
Voice metrics	Operational telemetry about latency, transport, and renderer behavior.
Voice envelopes	Renderer- and turn-local provenance, distinct from RFC-009 publication proof.
RAG and assistant corpus	Retrieval surfaces that support the interaction layer rather than normative audit truth.
Voice transport	STT/TTS/voice gateway surfaces that stay outside Core.

What Now Proxies to Core

Audit-facing LINA routes were not deleted as a namespace, but they were stripped of their long-term role as a second normative write surface. They now proxy into the live Core audit endpoints.

LINA route family	Current behavior
/api/lina/replay	Core audit replay
/api/lina/bundles/publish	Core audit publications
/api/lina/signatures/envelope	Core audit signature envelopes
/api/lina/registry/signer	Core signer registry

Interaction Model

The interaction promise is still the same from the user’s point of view: ask, inspect, open, compare, publish, or verify through one coherent assistant surface. The architectural difference is that consequential audit facts are now owned and validated in Core before LINA renders them back into a user-facing interaction.

User intent
  -> LINA route / interaction runtime
  -> Next.js proxy + auth forwarding
  -> AIJIM Core on managed runtime
  -> Supabase-backed audit / story truth
  -> typed response
  -> LINA explanation, UI action, or voice rendering

Why this matters

The phrase “LINA is the interface of the system” is still useful, as long as it is read operationally rather than architecturally. LINA is the interface humans meet; Core is the normative substrate that discharges invariant-bearing writes.