[EPIC] Telemetry & resource dashboard #122

Goal. A local, in-account telemetry & resource dashboard so that when a user deploys their vector DB (S3 Vectors + DynamoDB), they can see what was created and how their queries perform — with the exact look-and-feel of the dynavec landing page.

Chosen stack (decided)

• Frontend: Vite + React + TypeScript + Tailwind CSS + Recharts
  • React → largest contributor pool; Vite → fast, no SSR needed for a local tool; Tailwind → lets us reuse the landing-page brand tokens (coral #e8623b / warm ink, Inter + JetBrains Mono) for instant theme parity; Recharts → matches the benchmark charts.
  • Alternatives considered: Next.js (SSR overkill for a local dashboard), Streamlit (Python-fast but can't match the custom brand).
• Backend: FastAPI (Python), shipped as the dynavec[dashboard] extra. Reads the user's AWS account with server-side credentials (never exposed to the browser) and serves the built React bundle + a JSON telemetry API.
• Data: reuse DynamoDB for a telemetry events table (in-account, no new infra); resources via describe/list calls; cost via the existing benchmarks/cost_model.
• Launch: dynavec dashboard CLI builds/serves the UI + API.

Milestones

• Foundation: stack scaffold, Tailwind theme, CLI, backend API, telemetry model + client hook
• Resource views: buckets/indexes/tables, namespace explorer, cost panel
• Telemetry & eval: query analytics, per-query drill-down, recall@k/MRR/nDCG, faithfulness (LLM judge), trends
• Polish: security (local-only, server-side creds), dark mode/theme parity, docs

Child issues are labeled dashboard (+ telemetry where relevant). Comment to get assigned.