Skip to content
Routario Docs

Ask AI

Module · Stable

Send a prompt to an LLM and get a typed response (text/int/float/bool/JSON/enum); pass one or several images to describe or analyse them (vision); turn on web search to answer from current pages and get the sources.

Routes to the configured LLM via the routing table in AI Settings. Set response_format: json and supply a response_schema dict to extract structured data — the parsed object is at ai_parsed.. Use response_format: enum with response_options for classification. ai_model echoes which model was actually used. image takes one reference or a list (e.g. a Zendesk attachments list); image_mode together sends them in one call for one answer in text (three angles of one part), image_mode each asks once per image and returns descriptions as {name, text} — one image failing does not fail the others. web_search: true lets the model search the web while it answers and adds citations (the pages it cites, as {url, title}), search_queries and web_searched (false when it answered without searching); web_search_country (e.g. CZ) sets the search location. It needs a Microsoft Foundry or OpenAI model — a model that cannot search fails the step instead of answering from memory. This is the right tool for reasoning, translation, summarisation, and free-text extraction; prefer specialized modules for data operations (Contacts, geocoding, weather, OCR).

When to use

Section titled “When to use”

A single LLM turn with a typed answer: reasoning, translation, summarisation, free-text extraction (response_format: json + response_schema), or classification (response_format: enum). The answer flows into context for downstream steps to branch on or use. Turn web_search on when the answer depends on facts that change (a price, a fee, a deadline) and the sources matter.

When not to use

Section titled “When not to use”

To ask a human, use Ask a person. For multi-step autonomy that calls tools, use Run agent (this is a single stateless turn — none of your tools; web search is the only built-in one). For structured data operations prefer the specialized modules (Contacts, Geocode, Weather, OCR) rather than a hand-written prompt.

Inputs

Section titled “Inputs”

Configured per use: prompt, system, image, image_mode, web_search, web_search_country, model, fallback_1, fallback_2, timeout, response_format, response_options, response_schema, max_tokens.

Outputs

Section titled “Outputs”
  • ai_response
  • ai_parsed
  • ai_model
  • text
  • descriptions
  • citations
  • search_queries
  • web_searched

Auto-generated from the skill registry (load_skills()). Do not edit by hand.