Does Milvus support retroactive registration?

Not in Gateco v1. Milvus does not have a listing API that Gateco can use to enumerate existing vector IDs. Resources must be registered via the Gateco ingestion API or manually via the resource API.

Is Milvus keyword search ranked?

No. Milvus keyword search uses a like expression for substring matching. All matching documents get score 1.0. Use the vector or hybrid mode for ranked relevance.

How does Gateco handle Milvus partitions?

The collection_name in the search config identifies the Milvus collection. Partition filtering can be added as a policy condition if your Milvus schema uses partition keys as metadata fields that Gateco can evaluate.

Vector DBTier 2

Milvus

Policy-enforced retrieval on Milvus for billion-scale vector search workloads.

Search capabilities

Vector (ANN)	Keyword (ranked)	Keyword (filter)	Hybrid (native)	Grep (substring)	Grep (regex)

Milvus HNSW: <10ms p99 ANN. Policy overhead <25ms p95.

sidecar metadatainline metadata

Milvus is a distributed open-source vector database designed for billion-scale ANN search with rich indexing options (HNSW, IVF, DiskANN) and native hybrid search. Gateco connects to Milvus as a Tier 2 retrieval connector.

Milvus supports named vector fields and sparse vectors for native hybrid search. Gateco's hybrid mode uses Milvus's built-in hybrid search pipeline combining dense and sparse vectors.

Keyword search on Milvus is filter-only: it uses a payload like expression for substring matching with uniform score 1.0. There is no BM25 or TF-IDF ranking. For ranked keyword relevance at scale, pair Milvus (for ANN) with a separate OpenSearch connector.

Milvus does not have a vector listing API in Gateco v1, so retroactive registration is not available. Resources must be registered via the Gateco API or ingestion pipeline.

Sample search config

{
  "collection_name": "knowledge_base",
  "vector_field": "embedding",
  "output_fields": ["content", "classification", "department"]
}

Sample policy

{
  "name": "Engineering document access",
  "effect": "allow",
  "rules": [{
    "conditions": [
      {"field": "principal.department", "operator": "eq", "value": "engineering"},
      {"field": "resource.classification", "operator": "neq", "value": "restricted"}
    ]
  }],
  "selectors": [{"connector_type": "milvus"}]
}

Policy conditions reference resource.* and principal.* fields. Policy reference →

Frequently asked questions

Does Milvus support retroactive registration?: Not in Gateco v1. Milvus does not have a listing API that Gateco can use to enumerate existing vector IDs. Resources must be registered via the Gateco ingestion API or manually via the resource API.
Is Milvus keyword search ranked?: No. Milvus keyword search uses a like expression for substring matching. All matching documents get score 1.0. Use the vector or hybrid mode for ranked relevance.
How does Gateco handle Milvus partitions?: The collection_name in the search config identifies the Milvus collection. Partition filtering can be added as a policy condition if your Milvus schema uses partition keys as metadata fields that Gateco can evaluate.

Ready to connect Milvus?

Follow the step-by-step setup guide or talk to the team for help with your specific configuration.

Setup guide →Talk to the team All integrations