Embedding Resource

The embedding executor is a native capability compiled into the kdeps binary. It provides a SQLite-backed keyword store for indexing, searching, upserting, and deleting text documents. Use it as the storage layer for RAG pipelines that run fully on-prem.

Where it runs

Both workflow mode and agent mode. In workflow mode it executes as a DAG step. In agent mode, the workflow containing this resource runs as a single callable tool.

Configuration

# resources/store.yaml
embedding:
  operation: "index"                    # required: index | search | upsert | delete
  text: "document content here"         # required for index/search/upsert (optional for delete-all)
  collection: "default"                 # optional namespace (default: "default")
  dbPath: "/data/kdeps-store.db"       # optional path (default: "kdeps-embedding.db")
  limit: 10                             # optional max search results (default: 10)

Field	Type	Required	Default	Description
operation	string	yes	—	index, search, upsert, or delete
text	string	yes*	—	Text to index, query, or delete. *Optional for delete (omit to delete entire collection)
collection	string	no	"default"	Namespace for documents
dbPath	string	no	"kdeps-embedding.db"	SQLite database file path
limit	integer	no	10	Max results for search

Operations

Operation	Description
index	Insert text (ignored if duplicate)
upsert	Insert or replace text
search	Case-insensitive keyword search via LIKE
delete	Delete by text, or whole collection if text is empty

Output

Key	Type	Description
operation	string	The operation that was performed
collection	string	The collection used
success	bool	true on success
results	[]string	Matching texts (search only)
count	integer	Number of results (search only)
affected	integer	Rows deleted (delete only)
json	string	Full result as JSON string

Examples

Build a RAG pipeline

# Step 1: Scrape content
actionId: fetch
scraper:
  url: "{{ get('url') }}"

# Step 2: Index it
actionId: storeDoc
requires: [fetch]
embedding:
  operation: "index"
  text: "{{ output('fetch').content }}"
  collection: "knowledge"
  dbPath: "/data/store.db"

# Step 3: Search on user query
actionId: findDocs
embedding:
  operation: "search"
  text: "{{ get('query') }}"
  collection: "knowledge"
  dbPath: "/data/store.db"
  limit: 5

# Step 4: Answer with context
actionId: answer
requires: [findDocs]
chat:
  model: llama3.2:1b
  prompt: |
    Context: {{ output('findDocs').results }}
    Question: {{ get('query') }}
apiResponse:
  response: "{{ output('answer') }}"

Collections

Use collection to namespace documents — useful for multi-tenant or multi-topic stores:

# Index into separate collections
embedding:
  operation: "index"
  text: "..."
  collection: "contracts"

# Search only within one collection
embedding:
  operation: "search"
  text: "termination clause"
  collection: "contracts"

---

Note: This uses keyword (LIKE) matching, not vector similarity. For OpenAI vector embeddings, install the component:

kdeps registry install embedding

See Also

• Scraper Resource -- fetch content to index
• Search Resource -- search local files and the web
• LLM Resource -- use search results as context