Retrieval, fully managed

Managed Indexes

• HNSW, IVF-PQ and DiskANN
• Shard and replica planning
• Rebuilds with no read downtime
• Per-query latency headers

Nodeform picks HNSW under fifty million vectors, DiskANN above it, and moves you between them without a migration weekend. Rebuilds run on a shadow replica and cut over atomically, so reads never stop.

Hybrid Retrieval

• Dense vectors plus BM25
• Filters applied inside the index
• Reciprocal rank fusion built in
• One round trip, one response

Most teams bolt a keyword index next to a vector store and merge the results in application code. Nodeform runs both inside one engine and fuses them with reciprocal rank fusion, so filters narrow the search rather than throwing away its answers.

Private by Default

• Single-tenant indexes
• Encryption at rest and in transit
• SSO and audit logs included
• VPC peering or run-in-your-account

Every collection is single-tenant. Data is encrypted in transit and at rest, keys can be yours, and access is logged per request with the key that made it. Nothing is shared with another customer, including memory.

Embedding Sync

• Postgres, S3 and Kafka sources
• Bring your own model or ours
• Searchable seconds after a write
• Backfills without downtime

Every team that adopts vector search ends up building the same job: watch a source, embed the new rows, upsert them, handle failures, backfill after a model change. Nodeform runs it, with dead-letter handling and a backfill that never takes the index offline.