Wikipedia and Wikidata integration for AI Flowi Workflow

The Wikipedia API integration in AI Flowi Workflow includes Wikipedia and Wikidata direct components plus their separate API-tool variants. Direct components return Tables; the tool variants expose JSON and Tool ports. Select the source and output contract before configuring a query or connecting an Agent.

These four entries are documented in the catalogue; not production-tested by Ai Flowi. The reference uses AI Flowi Workflow 1.9.2, exported on 9 October 2026. It covers retrieval settings and output boundaries, without presenting an encyclopedia editing or administration interface.

By Ai Flowi · Published 10 October 2026

Official sources: MediaWiki API documentation; Wikidata documentation

Four entries for two retrieval sources

Use horizontal scrolling within the Wikipedia and Wikidata tables to compare the direct and tool interfaces.

The category index locates these identities. Display names alone are insufficient to decide the wiring: Wikipedia and Wikipedia API share input settings but expose different output types. The same distinction applies to Wikidata and Wikidata API.

Wikipedia inputs shared by direct and tool entries

tools.WikipediaAPI repeats this same five-field input contract. Only k is marked required. Set the query and language deliberately for a trial; optional flags do not establish useful behavior without text or identify which language is chosen when lang is omitted.

For the general count-versus-content distinction, see the search-limit explanation.

Wikidata has a separate query contract

The single required query input also appears on tools.WikidataAPI. Do not copy Wikipedia’s k, lang or metadata fields into it. The description says it performs a search; it does not establish SPARQL syntax, property traversal or a structured-statement export recipe.

Table, JSON and Tool are separate contracts

Both direct components name their Table port dataframe. Each tool entry has api_run_model as JSON and api_build_tool as Tool. Neither a Table nor JSON label defines article-text keys, entity identifiers, source URLs or reference fields. Inspect the representation from the selected component before creating field mappings.

Inspect Wikipedia and Wikidata separately

The Wikipedia and Wikidata contracts contain no login or key input; any surrounding access configuration belongs to the connections reference.

example configuration, not a recorded test

1. Add Wikipedia, identity wikipedia.WikipediaComponent. Set input_value to the illustrative topic Graph theory, required k to 2 and doc_content_chars_max to 1500.
2. Set lang to en as an illustrative language value to verify in your environment, and load_all_available_meta to true. The export supplies no language-code validation rule or metadata schema.
3. Connect Wikipedia’s dataframe to Chat Output‘s required input_value, which accepts Table. Inspect the actual content and available references before drafting anything from it.
4. For a separate source inspection, add Wikidata with query set to the same illustrative topic. Connect its dataframe to a separate Chat Output and review what entities or records are actually returned.
5. Keep the source identity with each observation. Matching query words do not prove that an encyclopedia page and a structured-data record represent the same intended subject.

Configure the two inspection views through the display reference while retaining their distinct Wikipedia and Wikidata source labels.

Use API-tool variants only after inspecting arguments

The Agent documentation supplies the receiving setup for either API variant’s api_build_tool output.

Inspect whether the Wikipedia callable can alter lang independently of its topic argument, then verify that its returned article concerns the mathematical subject in the example below.

What affects the quality and API scope of Wikipedia references?

Inspect ambiguous topics before accepting the retrieved subject. A short content limit can remove qualifiers; metadata can be absent even when requested. Preserve available source references and record which part of the content supports the intended use. A retrieved entry is background material, not independent confirmation of every statement it contains.

These inputs provide no edit, page-create, revision-history, login, namespace or administrative operation. The Wikidata query field also does not document full structured-data query support. Use an appropriate separate interface for those requirements instead of treating this component family as the entire MediaWiki API.

For synthesizing reviewed sources, use the research report writer guide. This page stops at retrieval contract inspection; it establishes neither current factual accuracy nor a complete source-verification workflow.

Related: arXiv integration for AI Flowi Workflow; Exa Search integration for AI Flowi Workflow; WolframAlpha integration for AI Flowi Workflow.

Decide whether the retrieved subjects match

Work through the illustrative Graph theory query by deciding what you want from each source. For Wikipedia, the question might be whether the returned article introduces the mathematical topic. For Wikidata, it might be whether an inspectable result identifies the corresponding subject. These are separate questions about the two returned representations. A shared query phrase is a starting point for comparison, rather than a join key that establishes equivalence between their records.

Inspect each candidate’s available description and reference before connecting the two observations. If an encyclopedia candidate concerns a person, book or software project with a similar phrase, do not pair it with a record for the mathematical subject. If Wikidata returns several plausible candidates, compare the identifying information actually supplied. When that information cannot resolve the ambiguity, leave the relationship unassigned. The query-only contract does not expose a property traversal or an entity-resolution operation to settle it automatically.

Language is a Wikipedia configuration decision in this export. In a second proposed article exercise, choose a language value that you have verified and keep the topic and count setting fixed. Inspect the subject again; matching labels across languages do not show that the returned pages contain the same wording or level of detail. Wikidata has no corresponding lang input here, so do not describe this exercise as a synchronized language switch for both sources.

For metadata inspection, first identify a concrete need, such as a source reference that lets a reviewer reopen the article. Then compare what is visible with load_all_available_meta disabled and enabled. If the desired reference remains absent, the enabled switch has not satisfied that need. Requesting all available metadata also does not establish a stable list of keys. Create a later mapping only for fields you can inspect in the selected result, and treat a missing optional field as missing.

When considering an API-tool variant, check how its argument contract handles topic and language together. An Agent able to change a topic might still rely on the node’s configured language; the export does not specify precedence. A tool call that retrieves a different subject cannot be rescued by fluent article-like prose. Resolve the subject match at retrieval time, then use the linked research guide if the task calls for synthesis across accepted sources.

Retrieval usage

Plans: Go from USD8/month billed yearly (USD10 monthly), Plus from USD16 (USD20 monthly), Pro 100 from USD80 (USD100 monthly), Pro 200 from USD160 (USD200 monthly) and Pro 500 from USD400 (USD500 monthly), with 5,000 to 250,000 Flowi Tokens/month and top-ups; 1 token per node run; third-party API and model usage is not included.

Explore further source categories through the integration directory.