API/AI Gateway/RAG Query & Document Ingestion

RAG Query & Document Ingestion

POST/v1/ai/{project_id}/rag

Ingest large documents (up to 5 MB) with configurable overlap and query context-grounded LLM responses.

rag.tstypescript

// 1. Ingest large text document into namespace
await aura.ai.ragIngest({
  document: 'Full text of enterprise security guide...',
  namespace: 'security-kb',
  chunk_size: 1000,
  chunk_overlap: 200,
})

// 2. Query grounded RAG response
const { data: answer } = await aura.ai.rag({
  question: 'How are database backups encrypted?',
  namespace: 'security-kb',
  system: 'Answer using only the provided context.',
})
console.log('Answer:', answer?.answer)
console.log('Sources cited:', answer?.sources)

200.jsonjson

{
  "data": {
    "answer": "Database backups are encrypted at rest using AES-256-GCM via Vault Transit KEK...",
    "sources": [
      {
        "id": "b1a2c3d4-...",
        "content": "All automated WAL and snapshot backups utilize AES-256-GCM keys...",
        "similarity": 0.912
      }
    ],
    "model": "claude-3-5-sonnet",
    "provider": "anthropic"
  }
}

Ingestion

Chunking Parameters

Parameter	Type	Required	Description
`document`	`string`	yes	Document content (up to 5 MB payload size).
`namespace`	`string`	yes	Target partition namespace.
`chunk_size`	`number`	no	Target characters per chunk (default `1000`).
`chunk_overlap`	`number`	no	Overlapping characters between consecutive chunks (default `200`).
`metadata`	`object`	no	Arbitrary JSON metadata attached to each chunk.

Management

Namespace Lifecycle Operations

GET /v1/ai/{project_id}/rag/namespaces: Lists all active RAG namespaces.
DELETE /v1/ai/{project_id}/rag/{namespace}: Deletes all chunks and vector entries in the namespace.
POST /v1/ai/{project_id}/rag/{namespace}/reindex: Recomputes embeddings for all chunks in the namespace when switching embedding models.

v1.0 · revised Oct 31, 2026
↗ OpenAPI spec↗ GitHub

← PREVIOUS

Semantic search

NEXT →

NL to SQL