Skip to content

STORES.md

Latest commit

 

History

History
98 lines (80 loc) · 4.15 KB

STORES.md

File metadata and controls

98 lines (80 loc) · 4.15 KB
Raw

Stores roadmap

Design doc: ../STORES.md

Adding the Org → Store → Documents layer. A store is the named collection you upload into and query against; it also carries the domain profile.

Current status (summary)

Phase Status
Phase 0 — Org → Documents (no stores) Shipped (today)
Phase 1 — CP + engine core (stores, store_id, default store) Not started
Phase 2 — dashboard (switcher, stores page, scoped views) Not started
Phase 3 — SDKs + MCP store support Not started
Phase 4 — profile-on-store, store-bound keys, per-store usage Not started

Everything is additive: a default store per org preserves today's single-pool behavior with zero breakage.

Phase 1 — control plane + engine core

The foundation. After this, every document silently belongs to a store; existing flows keep working via the default store.

  • CP migration: stores table (id, org_id, name, slug, description, profile, timestamps; UNIQUE(org_id, slug)). api_keys.store_id nullable FK. store_id on usage tables.
  • CP model: model/store.go; StoreID on APIKey + Principal.
  • CP store layer: Create / Get / GetByOrgSlug / ListByOrg / Update / Delete; EnsureDefaultStore(orgID).
  • CP handlers + routes: CRUD under /admin/v1/orgs/{orgId}/stores.
  • CP auto-default: call EnsureDefaultStore wherever orgs are ensured (signup / login / me), mirroring auto-org.
  • CP auth: APIKeyAuth resolves apiKey.StoreID; SessionAuth reads + validates X-Vectorless-Store against the org.
  • CP proxy: inject X-Vectorless-Store in proxy.Forward (resolution: header → key binding → org default).
  • Engine migration: documents.store_id + (org_id, store_id, created_at) index.
  • Engine db: Document.StoreID; GetDocument / ListDocuments / DeleteDocument / CountSections / LoadTree filter by store_id; NewDocument persists it.
  • Server handlers: requireStoreID + thread into documents / query / connect handlers.
  • Validate: upload with + without a store header; confirm default-store fallback; confirm cross-store isolation (store A can't see store B's docs).

Phase 2 — dashboard

  • lib/cp-proxy.ts: getActiveStore() + storeId option; inject X-Vectorless-Store.
  • StoreProvider (parallel to SessionProvider) + localStorage persistence of last-selected store.
  • StoreSwitcher dropdown in the sidebar header.
  • /dashboard/stores list + create + delete page; CP-proxy route(s) for store CRUD.
  • Scope existing views to the active store: documents list, upload, playground, analytics.

Phase 3 — SDKs + MCP

  • TS/Go/Python: storeId client config; createStore / listStores / getStore / deleteStore; per-call storeId override on ingest / query / list; transport sends the header (precedence: per-call > config > none).
  • MCP: mcp_store table + store_id on mcp_document; vectorless_create_store / _list_stores / _get_store / _delete_store tools; ingest/list/query accept + validate store_id.
  • Docs: refresh SDKS.md + MCP.md examples with stores.

Phase 4 — specialization + accounting

  • Profile on store: store declares its profile; CP passes X-Vectorless-Profile; ingest applies it. (Depends on the Profiles Phase 1 scaffold.)
  • Store-bound API keys: key creation UI + enforcement; store-bound key needs no header.
  • Per-store usage: surface usage rollups by store in the dashboard analytics.

Open questions (from design doc)

  • Cross-store query (default: no).
  • Move documents between stores (lean: immutable v1).
  • Slug vs id in dashboard URLs.
  • When per-store quota enforcement (vs. org-level) kicks in.

Related