Wikipedia Evidence Pack API for RAG

Give your AI system reusable, citation-ready evidence instead of raw search results. Each record includes a canonical Wikipedia URL, stable page ID, matched excerpt, freshness metadata, attribution details, and a normalized content hash.

Outcome

Turn a topic into citation-ready Wikipedia search records with stable page IDs, titles, descriptions, excerpts, source URLs and attribution.

Use cases

• Use the Actor when you need turn a topic into citation-ready Wikipedia search records with stable page IDs, titles, descriptions, excerpts, source URLs and attribution.
• Build workflows around normalized, deduplicable search evidence with stable IDs, ranking and explicit attribution — not a raw Wikipedia content mirror.
• Designed for AI application developers, RAG engineers, research teams, knowledge-base builders, and automation agencies.
• Schedule repeat runs: Scheduled monitoring detects revised pages and returns only new or changed evidence, reducing unnecessary embedding and indexing costs.
• Feed normalized dataset records into APIs, spreadsheets, databases, automations, or AI/agent workflows.

Quick start

1. Open the Actor input form.
2. Set the filters or query relevant to your workflow.
3. Start with a small maxResults value.
4. Run the Actor and inspect the default dataset before scheduling larger or recurring jobs.

Example input:

{
  "query": "retrieval augmented generation",
  "maxResults": 10
}

Inputs

• query (string, required) — Topic or phrase to search across English Wikipedia.
• maxResults (integer, optional) — Maximum number of evidence records to return. Values are capped at 100 per run.
• includeContent (boolean, optional) — Whether to fetch normalized page text or summaries for the selected results. The bounded search connector returns search evidence fields.
• includeRelated (boolean, optional) — Whether to add related or linked pages for broader RAG context. Related-page expansion is not included in the bounded search request.
• since (string, optional) — Optional ISO date or prior run timestamp used by downstream workflows for change detection.

Output example

The example below is taken from the latest validated local live-source sample when available; otherwise it is derived from the bundled dataset schema.

{
  "id": "75229858",
  "title": "Retrieval-augmented generation",
  "url": "https://en.wikipedia.org/wiki/Retrieval-augmented_generation",
  "description": "Type of information retrieval using LLMs",
  "excerpt": "<span class=\"searchmatch\">Retrieval</span>-<span class=\"searchmatch\">augmented</span> <span class=\"searchmatch\">generation</span> (RAG) is a technique that enables large language models (LLMs) to retrieve and incorporate new information from external",
  "relevance_rank": 1,
  "attribution": {
    "source": "Wikipedia",
    "license": "CC BY-SA 4.0",
    "license_url": "https://creativecommons.org/licenses/by-sa/4.0/"
  }
}

Pricing

Store pricing is configured in Apify. The intended primary paid unit is evidence record. Start with a small result limit while validating your workflow, then scale only when the output and economics fit your use case. The README never hard-codes a price that could drift from the live Store configuration.

Source and limitations

• Source: Wikimedia MediaWiki REST API for English Wikipedia.
• Licence/terms: Creative Commons Attribution-ShareAlike 4.0 International License (CC BY-SA 4.0); some content may have additional or different licensing terms.
• Reuse condition: Commercial reuse and redistribution of Wikipedia text are permitted when the applicable license is followed. Include attribution with a page hyperlink or stable URL, identify Wikipedia/Wikimedia as the source, preserve relevant license notices, and license modified or added Wikipedia text under CC BY-SA 4.0 or a compatible required license. This Actor must not imply Wikimedia endorsement. Images, logos, trademarks, imported text, fair-use material, and non-text media require separate license review. The public license permits commercial reuse, but it does not grant unrestricted rights to package all Wikimedia material as proprietary content or remove attribution/share-alike obligations. High-volume commercial access may require Wikimedia Enterprise or may be rate-limited.
• Automation condition: Use a descriptive User-Agent containing an application name and contact URL or email, obey Wikimedia robot and API etiquette policies, avoid abusive or disruptive request volume, implement throttling, retries, caching, and categorized handling for 429, 403, 404, and 5xx responses. Large-scale commercial use may be rate-limited.
• The product operates in a crowded generic RAG and web-research category, so positioning must emphasize Wikipedia-specific evidence packaging, attribution, freshness, and change detection.
• CC BY-SA permits commercial reuse but imposes attribution and share-alike obligations for reused or modified Wikipedia text; the Actor must not present source text as proprietary.
• Individual pages may contain imported material, fair-use content, images, or other elements with additional restrictions; default output should focus on text, metadata, and source URLs.
• Wikimedia may rate-limit high-volume commercial API traffic, and the public API is not a guaranteed high-throughput commercial data feed.
• Wikipedia content is community-edited and may be inaccurate, disputed, incomplete, or unsuitable as the sole source for high-stakes decisions.

This Actor does not bypass authentication, rate limits, robots/API policies, licensing conditions, or upstream access controls. Missing upstream values are not invented.

FAQ

Does this use the approved source contract?
Yes. Runtime requests are restricted to the source contract validated by Market Hunter before build.

Can I schedule repeat runs?
Yes, when the source permits automated access. Use conservative schedules and result limits that respect upstream conditions.

Why can some fields be empty or null?
The upstream source may omit fields for particular records. The Actor preserves missing values rather than fabricating data.

Troubleshooting

• If a narrow query returns no rows, loosen filters and retry with a small result limit.
• If the upstream source returns a rate-limit or temporary server error, retry later instead of increasing request pressure.
• If authentication or source identity is required, verify the protected credential/identity configured for the Actor.
• If the upstream response schema changes, report the failed run so the deterministic mapping can be updated.

Related Actors

• Wikidata Entity Enrichment API
• Wikipedia Page Change Monitor
• Wikipedia Citation and Reference Extractor
• Open Knowledge Graph Entity Resolver

Store publication remains an explicit manual action. This product is not affiliated with or endorsed by the upstream source unless the Store listing explicitly states otherwise.