The Lunaris AI Harness App — the graphical agent/conversation and observability surface for the AI layer. The full GUI door to the AI (the graphical counterpart to the terminal CLI harness): multi-turn conversation with streaming and visible tool calls, plus the read-only activity timeline (trigger → behaviour → predict → gate → act → audit) where the user reviews what the silent curator did and can undo individual curated actions.
It is not the autonomous trigger (the Event Bus is) and not the config surface (that is Settings → AI). This app is where you use and observe the AI.
Tauri 2 + SvelteKit + Svelte 5 + @lunaris/ui-kit. Built entirely on the
Lunaris Design System (single component source, no forked primitives) — a
greenfield exemplar of the unified UI.
Talks to: ai-daemon (query/chat, streaming), ai-agent over D-Bus
(org.lunaris.AIAgent1; agent state, undo/compensation), the audit ledger
(activity timeline, via audit-proto), and the Event Bus (live activity).
Read before building:
docs/architecture/ai-app.md— this app's plan (surfaces, transparency principles, the undo/compensation loop, the open-source GUI-client survey, build sequence A1–A6).docs/architecture/design-system.md— the UI canon this app is built on.docs/architecture/ai-agent-design.md— the AI layer, interaction model (silent curator + pull), capability model.
(Those docs live in the shared lunaris-sys/docs/ folder.)
Scaffold pending (build phase A1). This repo currently holds only the bootstrap.