Anvia

@anvia/neo4j API reference ​

Schema and extraction ​

ts
const schema = defineGraphSchema({ nodes, relationships })
const facts = await extractGraphFacts({
  model,
  schema,
  chunks,
  conflicts,
  instructions,
  retries,
  concurrency,
  abortSignal,
})

Import the provider-neutral defineGraphSchema() and extractGraphFacts() from @anvia/graph. defineNeo4jGraphSchema() and the adapter extraction export remain available for compatibility. Extraction returns { output, usage, warnings }, where output contains entities, relationships, and chunk-to-entity mentions; warnings reports fact conflicts that a conflicts policy resolved instead of rejecting.

Client and registrations ​

ts
class Neo4jClient implements AsyncDisposable {
  constructor(options: Neo4jClientOptions)
  managedKnowledgeGraph(options): ManagedNeo4jKnowledgeGraph
  tenant(tenantId): Neo4jTenant
  knowledgeGraph(options): Neo4jKnowledgeGraph
  nativeDriver(): Driver
  close(): Promise<void>
}

class Neo4jTenant {
  get namespace(): string
  managedKnowledgeGraph(options): ManagedNeo4jKnowledgeGraph
}

Managed registrations expose ensure(), replaceDocuments(), deleteDocuments(), retrieve(), and explore(). Existing registrations expose validate(), retrieve(), and explore() but no provisioning or mutation methods.

tenant() hashes the tenant ID into a sha256 namespace, and namespaces are created only through tenant(). A Neo4jTenant.managedKnowledgeGraph() provisions a tenant-scoped managed graph whose writes, vector and full-text search, traversal, evidence, and exploration are stamped and filtered by the __anvia_namespace property.

Retrieval and tool ​

ts
const context = await graph.retrieve({
  model,
  query,
  search,
  traversal,
  evidence,
  retries,
  abortSignal,
})

const tool = createGraphSearchTool({
  name,
  description,
  graph,
  model,
  search,
  traversal,
  evidence,
})

Import createGraphSearchTool() from @anvia/graph. Search is discriminated as vector or hybrid. Traversal requires an explicit schema relationship allowlist and bounds. Evidence is always explicit as { type: 'none' } or, for managed graphs, { type: 'chunks', maxChunks }.

Standalone retrieval and exploration ​

retrieveGraphContext() and exploreGraph() run the same policies without going through a registration method:

ts
const context = await retrieveGraphContext({
  graph,
  model,
  query,
  search,
  traversal,
  evidence,
  retries,
  abortSignal,
})

const view = await exploreGraph(graph, {
  mode: 'overview',
  includeProvenance: true,
  maxNodes: 100,
  maxRelationships: 200,
})

retrieveGraphContext() accepts the same search, traversal, and evidence policy as graph.retrieve(); evidence is { type: 'none' } for existing registrations and { type: 'chunks', maxChunks } for managed ones. exploreGraph() accepts the shared overview and expand options and returns nodes, relationships, and truncation flags; pass includeProvenance: true to attach GraphExploreProvenance source documentIds and chunkIds to nodes and relationships.

Ingestion and exploration ​

Use ingestGraphText(), ingestGraphDocuments(), and prepareGraphDocuments() from @anvia/graph. Both graph registration modes implement GraphExplorer:

ts
await graph.explore({ mode: 'overview', maxNodes: 100 })
await graph.explore({ mode: 'expand', nodeIds, maxDepth: 1 })

Public type families ​

Provider-neutral schema, fact, ingestion, retrieval, tool, exploration, and write contracts live in @anvia/graph. @anvia/neo4j exports Neo4j client, resource, registration, index, and compatibility types. GraphFactConflictError represents extraction conflicts that cannot be reconciled safely.

Built for Anvia.