`dodil k3 search`

Top-level vector search. A single command — text query only, with --collection, --top-k, --min-score. Maps to the Search RPC.

The API’s Search RPC is rich (three query shapes, three modes, multi-collection, rerank, pre-filter, fast-lane tuning). The CLI exposes only the text-query happy path. For everything else — hybrid + rerank, multi-collection, pre-filter, pre-embedded vector queries, multimodal — use the API. Gap callouts below.


dodil k3 search [query] -b BUCKET -c COLLECTION [-k TOP_K] [-s MIN_SCORE]

The query is a positional argument — quote it.

Flag	Short	Type	Default	Description
`--bucket`	`-b`	string	—	Required. Bucket to search in
`--collection`	`-c`	string	—	Collection name or ID. Omit to search all matching collections (the API’s multi-collection mode — but see gap below)
`--top-k`	`-k`	int32	`10`	Max results to return
`--min-score`	`-s`	float32	`0.0`	Minimum similarity score threshold

Examples

Simple text query against one collection:


dodil k3 search "what is multi-head attention" -b kb-prod -c docs

Top-5 results, JSON output for downstream processing:


dodil k3 search "transformer architecture" \
  -b kb-prod -c docs \
  -k 5 \
  -o json \
  | jq '.results[] | {score, object: .object.key, content}'

Filter by minimum similarity:


dodil k3 search "diffusion models" \
  -b kb-prod -c docs \
  -k 50 -s 0.7 \
  -o json

Multi-collection search via empty --collection — the API supports leaving collection_name empty to fan out across all compatible collections. The CLI accepts an empty -c but the practical reach is limited because the CLI doesn’t surface collection_statuses[] for per-collection observability. For real multi-collection workflows, use the API — see Recipes → Multi-collection Search.

CLI gaps — what’s on the API but not here

For these, drop to curl against POST /:bucket/vector/search — the API page shows the full pbjson body for each:

Feature	Use the API
Hybrid search (`SEARCH_MODE_HYBRID` / `_AUTO`) — dense + BM25 fusion via RRF	Search → Three search modes
Jina rerank (cross-encoder over top-K)	Search → Rerank
Pre-filter by metadata (FilterGroup with EQ/NEQ/GT/IN/CONTAINS/EXISTS, AND/OR groups)	Search → Pre-filter
Pre-embedded vector query (`VectorInput.values` or `.vectors[]` for batch)	Search → Shape 2
Multimodal file query (`s3_key` — image / audio / video)	Search → Shape 3
`include_content` / `include_highlights` — chunk text + highlight snippets in results	Search → Result shape
Milvus fast-lane tuning (`search_params.ef`, `nprobe`, `partition_names`, `output_fields`)	Search → Worked example: tuned HNSW

Common bridging snippet — when you outgrow the CLI

A typical “I want the CLI’s text-search flow but also need rerank + content”:


curl -sS -X POST "https://k3.dev.dodil.io/kb-prod/vector/search" \
  -H "Authorization: Bearer $DODIL_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "bucket": "kb-prod",
    "text": "what is multi-head attention",
    "collectionName": "docs",
    "topK": 5,
    "searchMode": "SEARCH_MODE_AUTO",
    "rerank": true,
    "includeContent": true
  }' | jq '.results[] | {score, object: .object.key, content}'

For everything else, see the full Search API reference — that’s the centerpiece page.

dodil k3 search

Examples

CLI gaps — what’s on the API but not here

Common bridging snippet — when you outgrow the CLI

See also

`dodil k3 search`