Skip to content

docs-architecture.md

Latest commit

 

History

History
40 lines (27 loc) · 2.04 KB

docs-architecture.md

File metadata and controls

40 lines (27 loc) · 2.04 KB
Raw

RUNE Architecture Notes

Principle

RUNE is an intelligence terminal. The core interaction is not chat; it is operating a live map of reality. Every surface should answer one of four questions:

  1. What is changing?
  2. Why is it changing?
  3. How unusual is the change?
  4. What is likely to happen next?

Runtime Layers

  • Terminal UI: Three-column desktop-first interface with responsive collapse for mobile.
  • Intelligence Domain: Strong TypeScript contracts for signals, forecasts, anomalies, and providers.
  • Provider Control Plane: BYOK registry with encrypted key ingestion, request normalization, prompt contracts, and model-agnostic routing.
  • Connector Layer: Source collectors normalize repositories, papers, markets, filings, social streams, and on-chain movement into observations.
  • Source/Data Plane: Supabase Postgres tables for topics, observations, scores, trend history, forecasts, and alerts.
  • Engine Pipeline: Collection, scoring, trend detection, anomaly detection, forecasting, confidence calculation.

Data Flow

connectors -> observations -> scoring -> anomalies -> triage -> forecasts -> snapshot/provider APIs -> terminal surfaces

Next Milestones

  • Persist provider configuration through Supabase RLS-protected tables.
  • Add source connectors and scheduled ingestion jobs.
  • Add D3 forecast confidence bands and React Flow clustering by topic category.
  • Add TanStack Query data hooks once API routes/server actions return persisted snapshots.

Provider Request Strategy

RUNE should not leak product behavior into provider-specific call sites. Provider adapters normalize request construction while the intelligence engine owns prompts, evidence packets, and confidence interpretation. This makes OpenAI-compatible providers, Anthropic, Gemini, Groq, and OpenRouter replaceable at the routing layer.

Operational Workflow

Anomalies should move through a small workflow: monitor, create watchlist, or escalate. This keeps RUNE from becoming another passive dashboard and turns signal detection into decisions.