Collections, Data & Index — Milvus-direct commands

The collection, index, data, and status commands operate on Milvus through your allocated database. They connect directly to the endpoint that dodil vbase db use wrote into your config — so run db use <service_id> first, or pass --db <name> per command.

These commands are convenience wrappers over standard Milvus operations. For the full Milvus data-plane surface — batch insert, filtered/hybrid search, partitions, queries — connect with the Milvus SDK against the same endpoint. See Recipes for end-to-end examples.

Every command below accepts --db <name> to target a database other than the active one, and -o json for machine-readable output.

Collections — `dodil vbase collection`

Command	Milvus op
`collection create <name>`	`CreateCollection`
`collection list`	`ShowCollections`
`collection show <name>`	`DescribeCollection` + `DescribeIndex`
`collection load <name>`	`LoadCollection`
`collection release <name>`	`ReleaseCollection`
`collection drop <name>`	`DropCollection`

`collection create`

Two schema modes.

Default mode — a primary-key field plus one vector field, configured by flags:


dodil vbase collection create products \
  --dim 768 \
  --id-type varchar --id-max-length 128

Flag	Default	Notes
`--dim`	`8`	Vector dimension. Must be > 0.
`--id-field`	`id`	Primary-key field name.
`--id-type`	`varchar`	`varchar` or `int64`.
`--id-max-length`	`64`	Max length for `varchar` ids.
`--vector-field`	`vector`	Vector field name.
`--vector-type`	`float_vector`	`float_vector` or `binary_vector`.
`--dynamic-fields`	`true`	Enable Milvus dynamic fields.

Custom schema mode — pass repeated --field specs in the form name:type[:pk][:key=value,...]. Exactly one field must be marked pk:


dodil vbase collection create products \
  --field "doc_id:varchar:pk:max_length=64" \
  --field "vector:float_vector:dim=768" \
  --field "category:varchar:max_length=32"

Supported type values: bool, int8, int16, int32, int64, float, double, varchar, json, float_vector, binary_vector. Vector fields require a dim= param; varchar defaults to max_length=64 if omitted.

Other collection commands


dodil vbase collection list
dodil vbase collection show products      # fields, shards, and indexes
dodil vbase collection load products      # load into memory before searching
dodil vbase collection release products   # free memory
dodil vbase collection drop products

A collection must be loaded before you can search it.

Index — `dodil vbase index`

Command	Milvus op
`index create <collection> <field>`	`CreateIndex`
`index drop <collection> <field>`	`DropIndex`


dodil vbase index create products vector --type HNSW --metric COSINE

Flag	Default	Allowed values
`--type`	`FLAT`	`FLAT`, `HNSW`, `IVF_FLAT`, `IVF_SQ8`
`--metric`	`L2`	`L2`, `IP`, `COSINE`

Choosing an index type

Type	Search	Strength	Trade-off
`FLAT`	Exact	Maximum recall; the correctness baseline	Latency grows fastest as data grows
`HNSW`	Approximate (graph)	Strong latency/recall default	Higher memory use
`IVF_FLAT`	Approximate (inverted file)	Scales well with tunable quality	Needs tuning for best recall
`IVF_SQ8`	IVF + quantization	Lower memory at large scale	More approximation loss

Choosing a metric

Metric	Better score	Use when
`L2`	Smaller	You care about absolute (Euclidean) distance.
`IP`	Larger	Both magnitude and direction matter (inner product).
`COSINE`	Larger	Angular similarity matters more than magnitude.

Set the index --metric to match how you intend to query. Note: the data search command always queries with metric_type=L2 (see below), so for first-class CLI search keep the index on L2, or run searches with a matching metric through the Milvus SDK.

Drop an index by field:


dodil vbase index drop products vector

Data — `dodil vbase data`

Command	Milvus op
`data insert <collection>`	`Insert`
`data search <collection>`	`Search`

`data insert`

Inserts a single row: one id and one float vector.


dodil vbase data insert products \
  --id doc-1 \
  --vector "0.12,0.04,0.91,0.33"

Flag	Default	Notes
`--id`	—	Required. The primary-key value (string).
`--id-field`	`id`	Must match the collection’s PK field name.
`--vector`	—	Required. Comma-separated floats.
`--vector-field`	`vector`	Must match the collection’s vector field name.

The CLI inserts the id as a varchar and the vector as a float_vector. For bulk loads, multi-field rows, or int64 ids, use the Milvus SDK.

`data search`

Runs a top-K vector search.


dodil vbase data search products \
  --vector "0.12,0.04,0.91,0.33" \
  --topk 5

Flag	Default	Notes
`--vector`	—	Required. Comma-separated query vector.
`--topk`	`10`	Number of results.
`--vector-field`	`vector`	Field to search against.
`--id-field`	`id`	Field returned alongside scores.


Result ID   Score
doc-1       0.030000
doc-9       0.142000

The command searches with metric_type=L2 and returns the id field plus the score. For other metrics, output fields, or filter expressions, search through the Milvus SDK.

Status & version


dodil vbase status     # Milvus CheckHealth against the active database
dodil vbase version    # CLI version + remote Milvus version


Status   Reason
true     []

status reports the health of the allocated database (Milvus), not the control plane — for the control-plane health probe see HealthCheck.

Collections, Data & Index — Milvus-direct commands

Collections — dodil vbase collection

collection create

Other collection commands

Index — dodil vbase index

Choosing an index type

Choosing a metric

Data — dodil vbase data

data insert

data search