Templates — API Reference

Package: dodil.k3.vector.v1 · Service: VectorService

ListTemplates here returns the vector-pillar template catalog — Scriptum templates whose category = embedding (server-pinned). Use the IDs returned here when calling AddVectorPipeline.

Same catalog, different filter as Pipelines → Templates and Tables → Templates. The Pipelines endpoint returns all templates; Vector pre-filters to category=embedding; Tables pre-filters to warehouse_compatible=true.

RPC	HTTP
`ListTemplates`	`GET /:bucket/vector/_templates`

gRPC setup — grpcurl, endpoints, reflection, and field-name casing — is covered once in Conventions → Using gRPC.

What’s in the vector catalog

Each vector template ships as an index + search pair — you AddVectorPipeline with the _index variant; the matching _search variant runs query-side automatically when you Search against the resulting collection.

Template (index)	Modality	Use case
`text_embedding_index`	text, pdf, docx, html, audio, video	The canonical RAG ingest. Chunks + embeds documents.
`code_embedding_index`	code (rust, python, javascript, typescript, go, java, c, cpp, …)	AST-aware chunking (tree-sitter) at function / class boundaries
`visual_embedding_index`	image, video, audio, pdf	Multimodal — frames / spectrograms / page renders
`face_embedding_index`	image	SCRFD detection + face embeddings
`object_embedding_index`	image	Open-vocabulary object detection (requires `labels` template-input)

Full descriptions live at Pipelines → Templates → The catalog.

`ListTemplates`

Request

HTTP


# All vector-pillar templates
curl -sS "https://k3.dev.dodil.io/kb-prod/vector/_templates" \
  -H "Authorization: Bearer $DODIL_TOKEN"
 
# Free-text search
curl -sS "https://k3.dev.dodil.io/kb-prod/vector/_templates?search=code" \
  -H "Authorization: Bearer $DODIL_TOKEN"
 
# Label filter — only image-modality templates
curl -sS "https://k3.dev.dodil.io/kb-prod/vector/_templates?labels[modality]=image" \
  -H "Authorization: Bearer $DODIL_TOKEN"

Response

HTTP


{
  "templates": [
    {
      "id": "text_embedding_index",
      "name": "Text Embedding Index",
      "description": "Chunk + embed text/PDF/docx/HTML/audio/video into a Milvus collection.",
      "tags": ["embedding", "text", "rag"],
      "labels": {
        "category": "embedding",
        "type": "text",
        "pipeline": "index",
        "role": "index",
        "modality": "text, pdf, docx, html, audio, video"
      },
      "category": "embedding",
      "modalities": ["text", "pdf", "docx", "html", "audio", "video"],
      "acceptedExtensions": ["pdf", "txt", "md", "docx", "html"],
      "acceptedContentTypes": ["application/pdf", "text/plain", "text/markdown"]
    }
  ]
}

Using a template — the canonical flow


# 1. Browse — pick a template
dodil k3 vector templates -o json | jq '.templates[] | {id, modalities}'
 
# 2. Inspect its contract (what inputs does it require?)
dodil k3 template get text_embedding_index -o json | jq '.contract.inputs'
 
# 3. Create a vector collection bound to it
dodil k3 vector collection add docs -b kb-prod \
  --template text_embedding_index \
  --description "RAG corpus"
 
# 4. Upload a document — K3's auto-generated ingest rule fires
dodil k3 object create ./paper.pdf -b kb-prod -k papers/paper.pdf
 
# 5. Search
dodil k3 search "your query" -b kb-prod -c docs

For the full recipe (including the auto-generated rule, pipeline binding, multi-document ingest), see Recipes → Pipeline Collection.

Templates — API Reference

What’s in the vector catalog

ListTemplates

Request

HTTP

gRPC

Response

HTTP

gRPC

Using a template — the canonical flow

See also

`ListTemplates`