feat(wasm): publish @ruvector/rabitq-wasm and @ruvector/acorn-wasm to npm

[ruvnet]

Summary

Two new WASM packages, both v0.1.0, scoped under @ruvector, mirroring the existing @ruvector/graph-wasm packaging pattern.

• @ruvector/rabitq-wasm@0.1.0 — RaBitQ 1-bit quantized vector index for browsers / Workers / Deno / Bun. Wraps the existing crates/ruvector-rabitq-wasm crate.
• @ruvector/acorn-wasm@0.1.0 — ACORN predicate-agnostic filtered HNSW. New crates/ruvector-acorn-wasm crate wrapping the ruvector-acorn algorithm crate (PR research(nightly): ACORN — predicate-agnostic filtered HNSW #391).

Both packages are already published to npm (the publish was the implementation step, not a follow-up).

ADRs

• ADR-161 — packaging rationale for @ruvector/rabitq-wasm
• ADR-162 — packaging rationale for @ruvector/acorn-wasm (new wrapper crate)

What ships in each package

Both follow the standard wasm-pack 3-target layout:

npm/packages/rabitq-wasm/        web target (default `import`)
npm/packages/rabitq-wasm/node/   nodejs target
npm/packages/rabitq-wasm/bundler/  bundler target (Vite, Webpack, Rollup)

Same for acorn-wasm/. Per-package files:

• package.scoped.json — canonical scoped package.json (committed). build.sh copies this over the package.json that wasm-pack regenerates on each build, so the published config is reproducible.
• README.md — install + usage examples for browser, Node, bundler.
• .gitignore — excludes the generated .wasm + .js + .d.ts so only canonical source lives in git.

The wasm-pack-generated artifacts are NOT committed (matches the existing graph-wasm convention).

API surface

@ruvector/rabitq-wasm

import init, { RabitqIndex } from "@ruvector/rabitq-wasm";
await init();

const idx = RabitqIndex.build(vectors, dim, /*seed*/ 42n, /*rerankFactor*/ 20);
const results = idx.search(query, /*k*/ 10);
//  → [{ id, distance }, ...]

@ruvector/acorn-wasm

import init, { AcornIndex } from "@ruvector/acorn-wasm";
await init();

const idx = AcornIndex.build(vectors, dim, /*gamma*/ 2);
const results = idx.search(query, /*k*/ 10, (id) => metadata[id].in_stock);
//  → [{ id, distance }, ...]

The acorn-wasm predicate is a JS callable invoked once per node visited (≤ ef ~150), so any closure works — equality, range, geo, ACL, composite — with no schema coupling.

Verification

• cargo check -p ruvector-acorn-wasm -p ruvector-rabitq-wasm clean
• cargo clippy -- -D warnings clean for both crates
• wasm-pack build succeeds for all three targets on both crates
• npm publish --access public succeeded for both — versions are live:
  • npm view @ruvector/rabitq-wasm version → 0.1.0
  • npm view @ruvector/acorn-wasm version → 0.1.0

Branch context

This branch carries cherry-picks of:

• a674d6eba (ruvector-rabitq-wasm crate from feature/rabitq-wasm) — the base crate the npm package wraps.
• b90af9caa, 0b4eab11f, eb88176bd, f5913b783 (the ruvector-acorn crate + optimizations from PR research(nightly): ACORN — predicate-agnostic filtered HNSW #391) — required to build the new ruvector-acorn-wasm wrapper.

When PR #391 lands, those four commits collapse to no-ops and this PR becomes purely additive: the new wrapper crate + ADRs + npm package wrappers.

Test plan

• cargo check --workspace includes both new crates
• cargo clippy -p ruvector-rabitq-wasm -p ruvector-acorn-wasm -- -D warnings clean
• bash crates/ruvector-rabitq-wasm/build.sh produces all three targets
• bash crates/ruvector-acorn-wasm/build.sh produces all three targets
• npm pack --dry-run shows the expected file list (6 files: README, package.json, .wasm, .wasm.d.ts, .js, .d.ts)
• npm publish --access public succeeded for both
• CI workspace shards stay green (the new crates are tiny WASM façades; check-wasm-dedup catches duplicate-export collisions)

Follow-up

• Wire wasm-pack build for both crates into a release-please workflow so future version bumps publish automatically.
• Consider an integration test crate that imports both WASM packages from npm and runs a smoke script.
• @ruvector/cognitum-gate-wasm directory exists but was never published — orphaned. Worth cleaning up or shipping in a separate PR.

🤖 Generated with claude-flow