Lexery Wiki

Lexery - Deployment and Infra

3 хв читання

Compiled from

  • raw/codebase-snapshots/supabase-schema-2026-04-09.md

Lexery - Deployment and Infra

Infrastructure Stack At A Glance

Observed building blocks

  • apps/portal: Next.js frontend
  • apps/api: NestJS control plane
  • apps/brain: Node legal runtime
  • apps/doclist-resolver-api: Cloudflare Worker
  • managed/storage layers across docs and code: Supabase, Qdrant, Cloudflare R2, OpenRouter

Current Infra Roles

Supabase

  • identity / auth integration
  • relational control-plane data
  • run and supporting DB surfaces in architecture docs

Qdrant

  • vector / semantic retrieval roles
  • present across legislation and memory architecture

Cloudflare R2

  • attachment overflow
  • run artifacts
  • legacy legislation canonical storage

OpenRouter

  • model routing layer for current AI calls in Brain and portal server routes

Cloudflare Workers

  • concrete current use in DocList resolver service

Azure Target

Observed in docs

  • 03-azure-deployment.md describes: api-gateway, agent-gateway, agent-worker
  • Uses: ACA, VNet, KEDA, Key Vault, ROLE=API and ROLE=WORKER

Drift warning

  • Observed: current checked code does not yet prove this deployment split end-to-end.
  • Best reading: Azure topology is a documented target architecture, not fully confirmed live runtime truth.

GitHub / CI Evidence

  • apps/portal/.github/workflows/ci.yml
  • .github/workflows/lldbi-brain-admin.yml

Meaning

  • infra and verification are already subsystem-specific
  • current project is not a single uniform deploy unit

Portal Deployment Hints

  • apps/portal/Dockerfile exists
  • Portal README assumes server-side environment and OpenRouter key usage

Legacy Infra Meaning

  • Bridge repo legislation work shows infra maturation before monorepo productization.
  • Lexery likely earned its current architecture through corpus and pipeline pain, not greenfield elegance.

Best Synthesis

Lexery infra is currently a hybrid stack:

  • product shell in modern web app form
  • legal runtime in Node service form
  • Cloudflare used concretely for data-plane utilities
  • Azure described as target hosting shape
  • managed data services used as domain scaffolding

Main Risk

Documentation about desired deploy topology is richer than hard proof of current deployed topology.

Топологія сховищ (за кодом)

У конфігурації Brain (apps/brain/lib/config.ts) та CI видно два логічні шари керованих сервісів — окремо для продукту та для законодавчого корпусу.

Supabase

  • Два проєкти (~2): реляційна Lexery legal-agent DB (SUPABASE_LEXERY_LEGAL_AGENT_DB_*) для runs, control plane, MM-таблиць тощо, і окремий Legislation (SUPABASE_LEGISLATION_*) для метаданих законодавства / taxonomy (опційно, з graceful degrade).
  • Детальніше про ролі: Lexery - Storage Topology.

Cloudflare R2

  • Два bucket за замовчуванням: legislation (канонічний JSON / LLDBI) vs lexery-legal-agent (runs, overflow, MM Docs сирі артефакти — через R2_RUNS_BUCKET / MM_DOCS_BUCKET).
  • Env aliases у коді: R2_LEGISLATION_BUCKET, LLDBI_R2_BUCKET, R2_BUCKET_RUNS.

OpenRouter

  • Єдиний ключ для LLM і embeddings у Brain: пріоритет OPENROUTER_API_KEY_BRAIN > OPENROUTER_API_KEY_ONLINE (getOpenRouterKeyFromEnv).

Qdrant

  • Legislation cluster — U4 law retrieval (QDRANT_URL / QDRANT_CLUSTER_ENDPOINT_LEXERY_LEGISLATION_DB), колекції на кшталт lexery_legislation_chunks / lexery_legislation_acts.
  • Lexery-LA cluster — семантична пам’ять (lexery_memory_semantic_v1) та MM Docs (спільний endpoint за замовчуванням, окремі колекції); fallback на legislation для memory лише за явним прапором MEMORY_QDRANT_ALLOW_LEGISLATION_FALLBACK.

Перехресні пояснення: Lexery - Retrieval, LLDBI, DocList, Lexery - Memory and Documents, Lexery - Provider Topology.

See Also

Вигляд графа

Зворотні посилання