[github-actions]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

@rushdb/javascript-sdk@3.0.0

Major Changes

• 98c2647: Direct mode for 3.0 rust native engine (rc)

Minor Changes

• 98c2647: Add direct mode for the standalone RushDB engine, including record replacement, deletion by ID, and single-record lookups.

• 98c2647: Direct mode (a standalone engine) and RushDB Cloud are one SDK with one code path.

  • In direct mode every method is answered from the engine: records (including find with
    orderBy, select and groupBy, upsert, set, findOne/findUniq, findById of several
    ids), relationships, labels, properties, JSON and CSV import, CSV export, text indexes and keyword
    search, external vector indexes, vector and hybrid search with a queryVector, raw openCypher, the
    schema and transactions. Nothing is ever sent to RushDB Cloud: before, a method without direct
    support sent the engine token to api.rushdb.com.
  • The Cloud-only features (managed embeddings, ai.search, relationships.patterns.*,
    settings.get) throw CloudOnlyError before sending anything. CLOUD_ONLY_METHODS,
    CLOUD_ONLY_FORMS and isCloudOnlyMethod list them.
  • In a transaction, direct mode stages record creates, update, set, deleteById, attach,
    importJson and importCsv (records are named by id; the engine resolves them inside the
    transaction). Writes the engine cannot stage (bulk deletes, detach, upserts and merge imports,
    inline vectors) throw DirectModeError instead of running outside the transaction; the engine
    refuses a merge import or inline vectors in a staged import.
  • records.detach without typeOrTypes removes relationships of every type, and without
    direction removes them both ways, as RushDB 2.x on Neo4j did. records.attach and
    relationships.createMany without a type create __RUSHDB__RELATION__DEFAULT__ relationships,
    as before, instead of RELATED_TO.
  • direct: true without engineUrl throws instead of falling back to RushDB Cloud. New scope
    option for a multitenant engine; the unused engineToken option is gone.
  • Fixed: direct mode never sent the request body, so every call failed against a real engine.
  • Fixed in both modes: records.findById([...ids]) posted to the create endpoint; it now searches
    by id.
  • The package declares exports: under Node, import gets an ES module build (dist/index.node.mjs)
    whose default export is the RushDB class; before, an ESM import RushDB from received the
    CommonJS module object.

• 98c2647: Hybrid search fusion controls. fusion: 'dbsf' fuses each leg's scores normalized by its own
  distribution (mean ± 3σ, as Qdrant defines DBSF) instead of ranks (rrf, the default, with
  rrfK). Each leg takes a weight (its share of the fused score) and a prefetchLimit (hits it
  ranks before fusion, 1–10,000). On vectorSearch, the top-level weight/prefetchLimit are the
  vector leg's and text.weight/text.prefetchLimit the text leg's; the result reports fusion.

• 98c2647: Embedding indexes have names. A name is unique within a project, defaults to
  <label>.<property>.<model>.<dimensions>.<similarity>, and is returned when indexes are listed.
  Search (index), inline vectors (index), stats, vector upsert and delete accept an index's name
  as well as its id. The dashboard asks for the name in a second step when creating an index,
  prefilled with the default.

• 98c2647: Keyword (BM25) search. Text indexes cover one property of a label, or every string of it (*),
  and are declared, listed and dropped through /text-indexes (db.textIndexes, MCP
  createTextIndex / listTextIndexes / deleteTextIndex), with a default name of
  <label>.<property>.text. POST /search/text (db.records.textSearch, MCP textSearch) ranks
  matching records among those labels and where select, and total counts every match. A
  vector search with a text leg becomes a hybrid search the engine fuses. The dashboard lists and
  creates text indexes on the Search Indexes page and adds a Keyword search mode to Records.

Patch Changes

• 98c2647: The RushDB engine is the only data store; what remained of the Neo4j-backed platform is removed.

  • Projects no longer carry a customDb (bring-your-own Neo4j) connection: the projects.custom_db
    column is dropped (migration 0010_drop_project_custom_db), GET /sdk/settings no longer
    returns customDb, DELETE /projects/:id no longer takes shouldStoreCustomDbData, and the
    dashboard drops the "External Instance" badge. Every project is metered the same way.
  • POST /query/raw returns { records } (the always-empty summary is gone). Raw queries are
    openCypher run by the engine, which must have the cypher language enabled.
  • A write sent with x-transaction-id defers its side effects (recount, schema refresh,
    relationship automation) to POST /tx/:txId/commit, and a failed request inside a transaction
    rolls the transaction back. Both checks keyed off a Neo4j transaction object that was no longer
    attached, so neither ran.
  • RUSHDB_TRANSACTION_TIMEOUT_MS no longer falls back to NEO4J_TRANSACTION_TIMEOUT_MS.
  • The CLI db-restore command now recounts the project after the restore; it never ran before.
  • Deploy templates (Compose, Helm, Fly) run the engine with --multitenant --languages cypher,
    which Core requires: it scopes every request to its project. Heroku's app.json sets
    RUST_ENGINE_URL / RUST_ENGINE_TOKEN, the variables Core reads.
  • SDK: serverSettings drops customDB / managedDB; doc comments no longer mention Neo4j.