dodil k3 vector collection

Manage vector collections. Five subcommands — add (pipeline-mode), add-manual (external-mode), get, list, delete. Mirrors Collections — API Reference.

Persistent flag on the whole group: --bucket / -b (required).

dodil k3 vector collection add — pipeline-mode

dodil k3 vector collection add [name] -b BUCKET -t TEMPLATE_ID [-d DESC]

Creates a template-driven collection. K3 spawns the index + search Scriptum scripts; the Milvus collection materializes lazily on first ingest. Schema (dimensions, metric, sparse mode, embedding type, embed_model) is owned by the template’s ScriptContract — caller cannot override.

# Text / PDF / docx → embeddings (canonical RAG)
dodil k3 vector collection add docs -b kb-prod \
  --description "RAG corpus" \
  --template text_embedding_index
 
# Code → embeddings (AST-aware chunking)
dodil k3 vector collection add code -b kb-prod \
  --description "Source-code search" \
  --template code_embedding_index
 
# Multimodal (images / video / audio / PDF page renders)
dodil k3 vector collection add assets -b kb-prod \
  --description "Mixed-media library" \
  --template visual_embedding_index

What K3 creates atomically:

1. The collection row (Milvus collection materializes lazily on first ingest)
2. A Scriptum pipeline bound to the index template
3. An auto-generated ingest rule with globs from the template’s acceptedExtensions

template_inputs are not on the CLI today. Some templates (e.g. object_embedding_index) require runtime inputs like labels. For those, use the API directly with the templateInputs field.

Browse available templates with dodil k3 vector templates — see templates.mdx.

dodil k3 vector collection add-manual — external-mode

dodil k3 vector collection add-manual [name] -b BUCKET \
  --dimensions N [--metric M] [--embed-model MODEL] [-d DESC]

Creates an external-mode collection — caller declares the schema, K3 creates the Milvus collection eagerly. No Scriptum binding; you push vectors via InsertVectors / UpsertVectors. The collection is recorded with embedding_source = EXTERNAL.

# OpenAI ada-002 shaped (1536 dims, cosine)
dodil k3 vector collection add-manual ada -b kb-prod \
  --dimensions 1536 \
  --metric cosine \
  --embed-model openai/text-embedding-ada-002
 
# Jina shaped (1024 dims)
dodil k3 vector collection add-manual jina-docs -b kb-prod \
  --dimensions 1024 \
  --embed-model jina-embeddings-v4

CLI gaps for add-manual:

• --embedding-type not exposed → CLI always creates EMBEDDING_TYPE_FLOAT collections. For binary / float16 / bfloat16 / int8 embeddings, use the API.
• --sparse-mode not exposed → CLI always creates SPARSE_MODE_NONE (dense-only). For BM25 or EXTERNAL sparse, use the API.

For the full breadth of add-manual (all five dense types + three sparse modes), see API Reference → Collections → AddVectorCollection.

dodil k3 vector collection get

dodil k3 vector collection get [collection_id_or_name] -b BUCKET

Returns the full collection row — status, schema, embedding source, pipeline binding (pipeline-mode), template inputs (pipeline-mode).

dodil k3 vector collection get docs -b kb-prod -o json \
  | jq '{
      name, status, dimensions, distanceMetric, embeddingType,
      sparseMode, embedModel, modality, embeddingSource,
      embedPipelineId, templateId
    }'

Useful for confirming the schema before doing direct vector writes (embedding_type and dimensions must match VectorRecord.dense.*).

dodil k3 vector collection list

dodil k3 vector collection list -b BUCKET

Lists every collection in the bucket.

dodil k3 vector collection list -b kb-prod -o json \
  | jq '.collections[] | {name, status, embeddingSource, dimensions, embedModel}'

dodil k3 vector collection delete

dodil k3 vector collection delete [collection_id] -b BUCKET

Removes the K3 collection row + the underlying Milvus collection. Does not cascade to pipeline-mode collections’ bound Scriptum pipeline or auto-generated ingest rule — clean those up via Pipelines → DeleteRule + DeletePipeline.

dodil k3 vector collection delete col_a1b2... -b kb-prod

For pipeline-mode collections, the standard cleanup is:

# 1. Capture the bound pipeline + auto-rule before deleting
COLLECTION_ID="col_a1b2..."
PIPELINE_ID=$(dodil k3 vector collection get "$COLLECTION_ID" -b kb-prod -o json | jq -r '.embedPipelineId')
RULE_ID=$(dodil k3 ingest list -b kb-prod -p "$PIPELINE_ID" -o json | jq -r '.rules[0].ruleId')
 
# 2. Delete the rule, then the pipeline, then the collection (or any order)
dodil k3 ingest delete "$RULE_ID" -b kb-prod
dodil k3 pipeline delete "$PIPELINE_ID" -b kb-prod
dodil k3 vector collection delete "$COLLECTION_ID" -b kb-prod

Common gotchas

See also

• Collections — API Reference — full surface incl. template_inputs, all dense types, all sparse modes
• dodil k3 vector templates — discover templates for add
• dodil k3 search — query collections you created here
• Core Concepts → Collection — both creation modes side-by-side
• Recipes → Pipeline Collection + External Collection — end-to-end worked examples