Search resources

Three components search: web_search finds pages on the web, search_local walks files on disk, and embedding ranks text by keyword in a local SQLite store. Each has its own reference page.

Works in workflow and agent mode.

Resource	Does	Reference
component: web_search	Query the web via DuckDuckGo, return titles, URLs, and snippets - a kdeps component, YAML only.	web_search
component: search_local	Walk a directory by filename glob and content keyword. Optional TF-IDF index for ranked results - a kdeps component, YAML only.	search_local

web_search

# resources/find.yaml
actionId: find
name: Find Pages
component:
  name: web_search
  with:
    query: "kdeps workflow engine"   # required

# resources/respond.yaml
actionId: respond
requires: [find]
apiResponse:
  response:
    results: "{{ output('find') }}"

The output is the search engine's readable text - titles, URLs, and snippets. Feed a URL into web_scraper for the full page.

In agent mode the same tool is the web_search built-in, and the loop caps web_search/web_scraper calls per user request (20 by default; /model tool set web-limit <n> changes it).

See also

• Web scraper - read the text of a URL the search returned
• Embedding component - SQLite keyword store for on-prem RAG
• LLM resource - use search results as context for chat resources
• Resources overview - all resource types