Pull from Neo4j / Aura
relate pull [--from FILE] --query Q [--param N=V] --database PATH--from FILE—KEY=VALUEenv file (NEO4J_URI,NEO4J_USERNAME,NEO4J_PASSWORD, optionalNEO4J_DATABASE). Same format as Aura Created credentials andneo4j-cli --env-out-file. Omit--fromto use those process environment variables (NEO4J_USERaliasesNEO4J_USERNAME).--query Q— required shape query (file path or inline). MustRETURNnodes, relationships, or paths — not scalars or aggregates.--param N=V— source query parameter (repeatable).--database PATH— destination directory or.rlbundle.:memory:is rejected.
NEO4J_URI is an http(s) Query API origin (e.g. https://host:7473); a
Bolt neo4j://host:7687 maps to http://host:7474, an Aura neo4j+s:// URI
to https://<host>/. The hosted demo's Query API is
https://demo.neo4jlabs.com:7473 (database companies).
The command talks HTTP Query API v2, upserts the subgraph (source-scoped
union on remote elementId; overlapping pulls reuse existing local
elements), then carries the source schema (ADR-0201): SHOW CONSTRAINTS /
SHOW INDEXES rows are replayed as standalone CREATE … IF NOT EXISTS
statements (uniqueness / existence / key / property-type constraints and
RANGE / TEXT / POINT indexes; constraint-backing, offline, LOOKUP, FULLTEXT
and VECTOR rows are filtered or skipped with a stderr note — schema replay
never fails a pull). It checkpoints, prints
pulled N new / M reused nodes, P new / Q reused rels (+ C constraints, I indexes; S schema skipped) to PATH, and exits.
Pulled values keep their Cypher type (ADR-0181): pull asks the source for
the typed Query API v2 JSON media type, so POINT, DATE, TIME,
DATETIME, ZONED DATETIME, DURATION, UUID, VECTOR, and every other
source type are stored as those types. A property the engine cannot
represent — a byte array (Base64), an Unsupported value, an unparseable
payload, or an untagged cell — fails the pull with exit 1 and a
typed wire decode failed: … error naming the element id and property key.
The full source-type → stored-type table is in docs/api/cli/spec.md.