Perceptron Data
Research Agent
One call searches the sources you pick (Google web and news, EU and German law texts, open-access research, Wikipedia, Stack Overflow, Hacker News and GitHub), reads the pages and PDFs and keeps the passages that match the question most closely. Each source gets an evidence level from fixed rules, figures are compared across independent sites, and old sources and wording such as “amended by” are flagged, so your model can judge how far to trust its context.
Run on Apify $15.00 per 1,000 questions + $0.00005 start fee per run
What you get
- Evidence level for every source, set by domain and source rules – official, academic, preprint, institution, reference, news, documentation, community or commercial
- Figures compared across independent sites: consistent, single source or conflicting, with the sentences side by side
- Freshness check: age of the cited sources and wording such as “amended by”, “repealed” or “postponed” – a page gets a date only when it states one for itself
- EU law from the EU Publications Office in its current consolidated version (24 languages) and German federal law from gesetze-im-internet.de – name a provision and it is ranked first
- Open-access papers read in full (top 3 per question), not just the abstract
- Ready-to-paste context with numbered citations, or one row per passage with a stable ID for vector databases
Use cases
- Grounding for AI agents and chatbots (MCP, LangChain, n8n, API)
- Compliance assistants that must quote the law text
- Checking figures before they reach a report
- Feeding a vector database with cited, deduplicated chunks
Example input
{
"query": "What obligations do providers of high-risk AI systems have under the EU AI Act?",
"sources": [
"web",
"law"
],
"preferEvidence": "official_first"
}Price
$15.00 per 1,000 questions + $0.00005 start fee per run
Billed by Apify per answered question plus a start fee per run (per GB of memory); each further Google search beyond the included one (news, legal-text search, second round) is billed separately. Questions without relevant results, answers without Google (status partial), skipped or failed questions and the example question prefilled in the form are not charged.
Weekly quality check
Every Sunday an automated watchdog runs the live actors on Apify with fixed inputs whose correct answers are known, and checks the content of the results — not only whether the run succeeded. The results are published here as they are, failures included.
Last check on 12 Oct 2026: 10 of 10 cases passed, 1 of them with a note. Checked weekly since 8 Oct 2026.
Sample against the original source: On 12 Oct 2026 we took 20 result rows and compared 116 field values one by one with the original source: 108 matched (93.1%).
FAQ
Is a language model involved?
No. Search, ranking and all checks follow fixed rules, and nothing is sent to an AI provider.
What is not charged?
Questions without relevant results, answers given without Google (status partial), skipped or failed questions and the example question prefilled in the form.
How is the quality checked?
Every Sunday a watchdog re-runs all ten published examples on the live actor and checks the results – for legal questions, for example, that the named provision is cited first from the law text itself. On 12 October 2026, 20 passages from these runs were compared with the pages they cite: 108 of 116 fields matched, and the misses are published with their causes.
What are the limits?
The checks are signals, not verdicts. Evidence levels come from domain rules, so a site missing from the lists falls back to commercial; a page gets a date only when it states one for itself. Pages that need JavaScript are skipped and listed with the reason; pages that refuse automated requests are read through Apify's Web Fetch (at most 2 per question, billed by Apify on your account). Google returns at most 10 results per search.