Artixles Developer dashboard

Artixles Evidence API

Research answers from millions of academic papers, with figures checked against their sources; claim checks; reference checks against the official record; and one search across the major paper databases. For your own app, paid per call from a prepaid balance.

Quickstart

  1. Create a key on the developer dashboard. It is shown once: store it as a secret (for example ARTIXLES_API_KEY). The first key comes with €1 of free usage.
  2. Send it with every request: Authorization: Bearer artx_live_…. Requests and responses are JSON. Call the API from your server, never from a browser: the key must stay secret.
  3. Answers and checks read papers and can take up to about 4 minutes. Set your client's timeout to 300 seconds. If you close the connection, the call stops; it is charged once the AI work has started.

Every response includes usage: { cost_eur, balance_eur, request_id } and the headers X-Artixles-Cost, X-Artixles-Balance and X-Request-Id. balance_eur is null when the call was refused before the balance was read (see Errors). Errors are { error: { code, message } }.

Endpoints

Research answer POST /answer
€0.12 (deep: €0.30)
curl https://www.artixles.com/api/v1/answer \
  -H "Authorization: Bearer $ARTIXLES_API_KEY" \
  -H "Content-Type: application/json" \
  --max-time 300 \
  -d '{"question": "Does intermittent fasting help weight loss more than calorie restriction?", "deep": false, "year_from": 2015, "peer_reviewed": true}'

Returns answer (Markdown with [n] citations), references [{n, title, authors, year, journal, doi, url, formatted}], checks {figures_found, figures_checked, corrected, not_confirmed} and deep. Figures are checked against their papers and corrected in place: figures_found is how many the answer has, figures_checked how many could be checked against their paper, corrected how many were rewritten to match it, not_confirmed how many the paper didn't confirm. Optional: citation_style (numbered, apa, harvard, chicago, mla, ieee, vancouver, leidraad), year_to. The answer may contain ```chart blocks (JSON bars whose cite is a reference number) and ```mermaid diagrams.

Claim check POST /claims
€0.02 per claim checked, at least €0.10 per call
curl https://www.artixles.com/api/v1/claims \
  -H "Authorization: Bearer $ARTIXLES_API_KEY" \
  -H "Content-Type: application/json" \
  --max-time 300 \
  -d '{"text": "Coffee raises blood pressure for good. Regular exercise lowers the risk of depression.", "max_claims": 5}'

Up to 15 claims (max_claims, default 15). Each claim gets a verdict (supported, partly_supported, contradicted, no_evidence), the evidence with [n] citations, its sources and the best one to cite. You pay for the claims actually checked, at least 5 (€0.10) per call that ran, also when the text has fewer claims: the most it can cost is reserved up front and the rest is refunded in the same call.

Reference check POST /references
€0.005 per reference, €0.03 when its citing sentences were checked
curl https://www.artixles.com/api/v1/references \
  -H "Authorization: Bearer $ARTIXLES_API_KEY" \
  -H "Content-Type: application/json" \
  --max-time 300 \
  -d '{"references": ["Kahneman, D., & Tversky, A. (1979). Prospect theory. Econometrica, 47(2), 263-291."], "citing_text": "Losses loom larger than gains (Kahneman & Tversky, 1979)."}'

references: a list of strings (up to 60) or the pasted reference list as one string. For each: exists, details_match (with the official record), retracted, and, only when citing_text is given, citing_sentence_supported (true, false, or null when it couldn't be checked, e.g. a paywalled paper: that is neutral, never an error). The higher price applies only to references whose citing sentences were really checked against the paper; one that couldn't be checked (paywalled, no open text, not cited) costs the plain price. The price for the references the list holds is reserved up front and what isn't checked is refunded in the same call.

Search POST /search
€0.004 per query
curl https://www.artixles.com/api/v1/search \
  -H "Authorization: Bearer $ARTIXLES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "sleep deprivation working memory", "limit": 20, "year_from": 2018}'

Merged, de-duplicated papers from OpenAlex, Crossref, PubMed, Europe PMC, arXiv, DOAJ and OpenAIRE. Each paper has license, abstract (only where the article's licence allows reuse: CC BY, CC BY-SA, CC BY-ND, CC0, public domain; or from arXiv and DOAJ; otherwise null), summary (1–2 sentences in Artixles' own words from the abstract; null for CC BY-ND and CC BY-NC-ND articles, which allow no derivative works), checked_findings (figures and findings Artixles has checked against the paper), open_access (true, false or null when unknown) and oa_url. The response's attribution lists the credit lines the sources ask for: keep them with the data. limit 1 to 25; summaries: false skips the summaries. A search seen before answers in under a second; a new one asks every database and can take 30 seconds or more.

Usage and balance GET /usage
free
curl https://www.artixles.com/api/v1/usage -H "Authorization: Bearer $ARTIXLES_API_KEY"

Your plan, balance (plan allowance, top-up, trial), period end and this month's calls and spend per endpoint.

Prices

Fixed prices per call, paid from a prepaid balance. A call is only charged when it delivers: a call that fails on our side is refunded in full. Closing the connection (or cancelling) after the AI work has started is charged, because that work was done. A standard answer reads up to €0.05 of AI work and a deep answer up to €0.12, so prices never change with the question. All API prices are excl. VAT: VAT for your country is added at checkout, where businesses can enter a VAT number.

Research answer€0.12
Deep answer (more searches, more full texts)€0.30
Claim check, per claim checked (at least 5 per call: €0.10)€0.02
Reference check, per reference€0.005
Reference check with the citing-sentence check, per reference whose citing sentences were checked€0.03
Search, per query€0.004
Usage and balancefree

Plans

API plans are separate from Artixles chat plans: same account, its own balance. Each plan includes an amount of usage every month (below). Unused included usage carries over to the next month once. Top-ups (for subscribers) never expire and are used after the included usage.

Developer
€9 / month excl. VAT
€5 usage included
10 requests/minute per key
Pro
€19 / month excl. VAT
€15 usage included
30 requests/minute per key
Scale
€49 / month excl. VAT
€45 usage included
60 requests/minute per key
Business
€99 / month excl. VAT
€95 usage included
120 requests/minute per key

Top-ups (excl. VAT): €10 → €8 · €25 → €20 · €50 → €40 · €100 → €80 of usage, or any whole amount from €10 to €500. Free trial: €1 once, 10 requests/minute; after that, calls need a plan.

Cost estimator

Usage: €32 per month
Cheapest: Pro at €40.25/month excl. VAT (plan + about €21.25 of top-ups)

MCP server (Claude, ChatGPT, Cursor)

The same calls as tools for AI assistants, over the Model Context Protocol: search_papers, research_answer, check_claims and check_references. Same key, same prices, same balance. Server URL: https://www.artixles.com/api/mcp (Streamable HTTP). Add your key as a header, for example in Claude Code:

claude mcp add --transport http artixles https://www.artixles.com/api/mcp \
  --header "Authorization: Bearer $ARTIXLES_API_KEY"

Or in a client's JSON settings (Cursor and others):

{
  "mcpServers": {
    "artixles": {
      "url": "https://www.artixles.com/api/mcp",
      "headers": { "Authorization": "Bearer artx_live_…" }
    }
  }
}

research_answer and check_claims take 1 to 4 minutes: give the client a tool timeout of at least 5 minutes. Clients that accept Server-Sent Events get the tool result as an SSE stream with a keep-alive every 10 seconds, and progress notifications when the request carries a progressToken. Closing the connection stops the call (charged once the AI work has started). notifications/cancelled also stops it, but only on a best-effort basis: the server keeps no sessions, so a cancel that reaches another server instance is not seen.

Errors and limits

400invalid_request, invalid_json, request_too_largeA field is missing, has the wrong type or is too long. Nothing is charged; balance_eur is null.
401invalid_api_keyNo key, or the key is wrong or revoked. balance_eur is null.
402insufficient_balance, key_monthly_capThe balance (or the key's own monthly cap) can't cover the call. Refused before any work; includes top_up_url (and, for insufficient_balance, price_eur and the balance).
403account_blockedThe API account is paused. balance_eur is null.
429rate_limitedToo many requests this minute for this key. See Retry-After. balance_eur is null.
500internal_errorSomething failed on our side. A call that failed on our side is not charged. A call whose connection you closed (or that you cancelled) after the AI work started is charged; before that it is refunded.
503busy, paused, not_configuredArtixles is very busy, in maintenance or the API isn't available on this server. Not charged; retry later. balance_eur is null, except for busy.

Limits: question up to 4,000 characters; claim-check text up to 20,000; up to 60 references per call; request body up to 300 KB. Rate limits are per key per minute (X-RateLimit-Limit and X-RateLimit-Remaining headers). Each key can have its own monthly spending cap.