Epic: graduate the data layer to @gemstack/orm (+ adapters)

[suleimansh]

Summary

@gemstack/ai-sdk proved the graduation playbook: an engine matures inside a framework, proves it is framework-agnostic, then graduates to @gemstack/* while a thin binding stays behind. Architecture.md already ranks what comes next, and the data layer is the keystone:

@universal-orm/core (+ /drizzle / /memory / /rudder) -> @gemstack/orm (+ adapters). The ORM analog of @gemstack/ai-sdk. Mature, clearly agnostic. The strongest next candidate; move the core + its adapter family together.

Realized, GemStack becomes the unified home for agnostic engines: @gemstack/ai-sdk (AI), @gemstack/orm (data), @gemstack/schema (schema). This epic tracks the data-layer graduation. It is design-ready but execution-gated (see "The gate" below); nothing here lands before the brand-consolidation call with the Vike team.

The candidates (verified in vike-data today)

All four ORM packages are v0.1.0 and carry zero Vike imports (confirmed in their package.json deps + peers), which is exactly why they are the real candidates and the vike-* bindings are not.

Today	Graduates to	Notes
@universal-orm/core	@gemstack/orm	The engine. Agnostic, mature.
@universal-orm/drizzle	@gemstack/orm-drizzle	Adapter. Moves with the core.
@universal-orm/memory	@gemstack/orm-memory	Adapter. Moves with the core.
@universal-orm/rudder	@gemstack/orm-rudder (or stays @rudderjs)	Adapter; decide whether the Rudder adapter graduates or stays as the framework binding.
@vike-data/universal-schema	@gemstack/schema	Separate, lower-priority graduation. Agnostic but mis-scoped under @vike-data. Tracked here for sequencing only; can be its own epic.

The gate (the actual blocker, not a technicality)

Unlike ai-sdk — which graduated cleanly out of Rudder (a repo we own) — the ORM lives in the shared vike-data orbit and already ships under its own deliberate npm scope, @universal-orm. So this is not "where should orphaned code live." It is a brand question, quoting Architecture.md:

There are three agnostic-ish scopes in play (@gemstack, @universal-orm, @vike-data). The decision is not "where should this code live" but "do we consolidate the agnostic engines under one umbrella (@gemstack), or keep @universal-orm as a parallel brand?" Since @universal-orm is co-developed, this is a decide-with-the-Vike-team call, gated on brand traction.

Decision owners: GemStack + Vike maintainers, together. Not unilateral.

Three options to put to the Vike team:

1. Consolidate — fold @universal-orm/* into @gemstack/orm*; one umbrella for all agnostic engines. (Cleanest story; the thesis of Architecture.md.)
2. Parallel brands — keep @universal-orm as its own brand alongside @gemstack; document them as sibling agnostic-engine families.
3. Defer — hold until brand traction is higher (docs site live + gemstack.land cutover Attach gemstack.land custom domain to the docs site #63 is the signal building now).

Graduation playbook (only after the gate clears)

Follow the @gemstack/ai-sdk path exactly:

1. Copy @universal-orm/core source into packages/orm; rename to @gemstack/orm. Repeat for each adapter.
2. Reset to a fresh 0.x line.
3. Leave a re-export at each old name. Each old-name package takes one of two shapes (per Architecture.md):
   • Deprecated shim — pure re-export, slated for eventual removal (likely the core / drizzle / memory names).
   • Living framework binding — re-exports the agnostic core and keeps framework-coupled pieces that cannot graduate (e.g. the Rudder adapter's provider wiring / CLI), like @rudderjs/ai. Do not mislabel a binding as deprecated.
4. Repoint internal dependents (vike-data ORM bindings, @rudderjs/orm, playground twins) to @gemstack/orm*.
5. Update Architecture.md (move ORM from "candidate" to "shipped"), the root README package table, and the docs site.

Child issues (work top-down; #66 gates the rest)

• [orm #1] Decision: brand-consolidation with the Vike team (gate for #64) #66 — Decision: brand-consolidation with the Vike team (the gate)
• [orm #2] Publish @universal-orm/core + universal-schema (core, together) #67 — Graduate @gemstack/orm + @gemstack/schema (core, together)
• [orm #3] Publish @universal-orm/memory + @universal-orm/drizzle (adapters) #68 — Graduate @gemstack/orm-memory + @gemstack/orm-drizzle (adapters)
• [orm #4] Decide + execute the Rudder adapter shape #69 — Decide + execute the Rudder adapter shape
• [orm #5] Shims / living bindings at the old @universal-orm/* + @vike-data/universal-schema names #70 — Shims / living bindings at the old names
• [orm #6] Repoint dependents to @gemstack/orm* #71 — Repoint dependents to @gemstack/orm*
• [orm #7] Docs: move ORM from candidate to shipped #72 — Docs: candidate -> shipped

Schema folded into #67 (not a separate later item) per the Phase 0 spike (#65): the ORM test suite couples to schema, so they graduate together.

Out of scope

• @vike-data/kit does not graduate — it is the kit for authoring bindings, so it belongs with the binding ecosystem, not the engine umbrella (Architecture.md).
• Any vike-* binding package. Bindings stay with their framework and consume the engine.

Decision needed before any code lands

Pick option 1 / 2 / 3 above with the Vike team. Until then this epic stays design-ready, execution-gated — the queued plan, ready the moment the brand call is made.

[suleimansh]

Phase 0 spike done: #65 (draft, not for merge, nothing published).

Copied core + memory + drizzle + schema into the gemstack workspace as @gemstack/orm / -memory / -drizzle / @gemstack/schema, preserved their JS-src-direct build model, and got 154 tests passing / 0 fail with zero framework coupling.

One finding changes this epic: schema is not a "separate, lower-priority" graduation. @gemstack/orm's own test suite (and orm-memory's) builds fixtures with defineSchema/mergeSchemas, so schema must graduate with the ORM, not after it. The runtime is schema-free; the tests are not. I'll fold @gemstack/schema into the core graduation scope above rather than tracking it as a downstream item.

The Rudder adapter (@universal-orm/rudder, peer @rudderjs/database) was correctly excluded as the framework-coupled 'living binding' piece.

Everything past copy+rename (fresh 0.x, shims, repointing dependents, publish) stays gated on the brand-consolidation decision with the Vike team. #65 is the concrete artifact for that conversation.

[suleimansh]

Direction change from #66: the universal engines are not user-facing, so they stay under @universal-orm and do NOT graduate to @gemstack/*. The data layer ships as @universal-orm; GemStack remains the user-facing umbrella (AI family + MCP).

Knock-on:

• [orm #2] Publish @universal-orm/core + universal-schema (core, together) #67 / [orm #3] Publish @universal-orm/memory + @universal-orm/drizzle (adapters) #68 re-scope from 'rename to @gemstack' to 'publish under @universal-orm' (if we publish them public at all).
• PR spike: graduate the data layer to @gemstack/orm (+ adapters + schema) [#64 Phase 0, not for merge] #65's private copies on gemstack main stay dormant / unpublished (do not publish under @gemstack).
• Docs ([orm #7] Docs: move ORM from candidate to shipped #72 / PR docs: add the data family (orm + schema + adapters) [#72, HOLD until publish + #66] #73) rework so the data layer is not presented as a GemStack family.
• [orm #4] Decide + execute the Rudder adapter shape #69 (Rudder binding shape) is unaffected.
• SYNC RULE unchanged: vike-data stays the single source of truth for the engines.