API Reference

Base URL: https://api.factivelabs.com

All endpoints require Bearer token authentication unless noted. Responses are JSON.

POST /api/v1/verify

Submit content for fact-checking. Extracts individual claims and verifies each one against web sources. Supports three response modes: synchronous, streaming (SSE), and asynchronous (job polling).

POST /api/v1/verify Verify content for factual accuracy

Request Body

ParameterTypeRequiredDescription
content string Conditional Raw text to verify. Required when content_type is text, html, or any of the social/AI-paste types (twitter, reddit, instagram, chatgpt, perplexity, gemini, claude, gist). Max 250K characters.
url string Conditional URL to fetch and verify. Required when content_type is url, youtube, or tiktok.
file string Conditional Base64-encoded file content. Required when content_type is pdf, docx, doc, rtf, or image.
content_type string Optional Input format. Default text. Full set: text, html, url, youtube, tiktok, pdf, docx, doc, rtf, image, twitter, reddit, instagram, chatgpt, perplexity, gemini, claude, gist. See the Input Formats guide for which field (content/url/file) each type expects.
mode string Optional "claim" (single claim) or "document" (extract all claims). Auto-detected by length if omitted.
stream boolean Optional Return SSE event stream instead of JSON. Default: false.
async boolean Optional Return job ID immediately and poll for results. Default: false.
max_claims integer Optional Maximum claims to extract and verify (1–200). Default: 50.
context string Optional Additional context to improve verification accuracy (max 5,000 chars).
skip_table_claims boolean Optional Auto-skip granular table/statistical claims that can't be web-verified. Saves cost. Default: true.
second_pass boolean Optional Enable Sonnet second-pass verification for disputed claims in time-sensitive categories. Default: false.
include_sections array[int] Optional Section indices to fact-check (from /api/v1/analyze-structure). Mutually exclusive with exclude_sections.
exclude_sections array[int] Optional Section indices to skip. Mutually exclusive with include_sections.
webhook_url string Optional If async mode, POST results to this URL when complete.
callback_metadata object Optional Opaque JSON echoed back unchanged in the response.
profile string Optional Pipeline profile override. Most callers should leave this unset and let the API auto-select. Available profiles change over time; ask your account contact before pinning one.
extract_only boolean Optional If true, run extraction only and skip verification. Returns claims with verdict: null. Cheaper than /verify + faster than calling /api/v1/extract separately when you already have a /verify client. Default: false.
claims array[object] Optional Pre-extracted claims to verify directly — skips the extraction stage. Each item must have text (the claim) and may optionally include sentence, start_offset, end_offset. Useful when you already have claims from a prior /verify call with extract_only: true.
use_private_corpus boolean Optional Route claims through the customer's private corpus instead of the public web. Requires documents uploaded via POST /api/v1/private-corpus/upload. Default: false.
corpus_mode string Optional How private-corpus results combine with public web sources when use_private_corpus: true. Values: "corpus_only" (default — corpus only), "corpus_and_web" (corpus first, fall back to web on inconclusive).
corpus_scope string Optional One-paragraph description of what the corpus contains (max 2,000 chars). Improves corpus-routing accuracy when use_private_corpus: true. If omitted, the API uses the auto-generated scope from GET /api/v1/private-corpus/scope.
exclude_domains array[string] Optional Domains to exclude from web-search retrieval (e.g. ["reuters.com", "infowars.com"]). Bare hostnames only — subdomains are excluded automatically. Cap: 100 entries per request. Note: Gemini-grounded fallback paths can’t honor this filter; claims that fall through to fallback may still cite blacklisted sources. See Source Blacklist.
end_user_id string Optional Sub-tenant identifier (max 128 chars). Use to isolate your end-users’ data under your single API key. See Sub-tenanting.

Response (Synchronous)

{
  "id": "fc_abc123def456",
  "status": "complete",
  "input_type": "text",
  "title": "",
  "extracted_text": "The Earth is approximately 4.5 billion years old.",
  "claims": [
    {
      "text": "The Earth is approximately 4.5 billion years old",
      "sentence": "The Earth is approximately 4.5 billion years old.",
      "sentence_text": "The Earth is approximately 4.5 billion years old.",
      "start_offset": 0,
      "end_offset": 49,
      "span_start": 0,
      "span_end": 49,
      "verdict": "confirmed",
      "summary": "Confirmed by radiometric dating evidence.",
      "explanation": "Scientific evidence from radiometric dating consistently supports...",
      "categories": ["science"],
      "verified_by": "Sonar",
      "skipped": false,
      "context_flags": {},
      "unclear_reason": null,
      "sources": [
        {
          "title": "Age of Earth - USGS",
          "url": "https://www.usgs.gov/...",
          "domain": "usgs.gov",
          "publisher": "USGS",
          "snippet": "The age of the Earth is estimated at 4.54 billion years...",
          "date": ""
        }
      ]
    }
  ],
  "counts": {
    "total": 1,
    "confirmed": 1,
    "disputed": 0,
    "inconclusive": 0,
    "skipped": 0
  },
  "highlights": [...],
  "processing_time_ms": 3200,
  "usage": {
    "claim_count": 1,
    "skipped_claims": 0,
    "cost_usd": 0.01
  }
}

Response (Async)

HTTP 202 Accepted

{
  "id": "fc_abc123def456",
  "status": "queued",
  "created_at": "2026-05-14T10:30:00Z",
  "message": "Job queued. Poll GET /api/v1/jobs/{id} for results."
}

SSE Events (Streaming)

event: claim_verified
data: {"claim": {"text": "...", "verdict": "confirmed", "summary": "...", "explanation": "...", "skipped": false, "context_flags": {}, "unclear_reason": null, "sources": [...]}, "skipped": false}

event: highlights
data: {"highlights": [...], "source_length": 1234, "source_hash": "a1b2c3d4e5f6"}

event: complete
data: {"id": "fc_abc123", "status": "complete", "title": "", "claims": [...], "counts": {"total": 3, "confirmed": 3, "disputed": 0, "inconclusive": 0, "skipped": 0}, "usage": {"claim_count": 3, "skipped_claims": 0, "cost_usd": 0.03}}

POST /api/v1/verify/paragraph

Fact-check one paragraph of text in a single stateless call. Each request is fully self-contained — no session, no shared state. The canonical endpoint for streaming-style integrations: call it once per \n\n-separated paragraph in your LLM's output.

POST /api/v1/verify/paragraph Stateless single-paragraph verify (JSON)

Request Body

ParameterTypeRequiredDescription
text string Required One complete paragraph of text. Max 20,000 characters. Typically the output between two \n\n breaks of an LLM stream.
context string Optional Surrounding context to disambiguate references (e.g. the user's question or the previous paragraph). Max 5,000 characters.
profile string Optional Pipeline profile override. Leave unset for auto-selection.
use_private_corpus boolean Optional Route claims through your private corpus instead of the public web. Default: false.
corpus_scope string Optional One-paragraph description of what the corpus contains (max 2,000 chars). Improves corpus-routing accuracy.
second_pass boolean Optional Enable Sonnet second-pass verification for disputed claims. Default: false.
max_claims integer Optional Max claims to extract from this paragraph (1–200). Default: 50.
exclude_domains array[string] Optional Domains to exclude from web-search retrieval. Same semantics as on POST /api/v1/verify. Cap: 100 entries per request. See Source Blacklist.
end_user_id string Optional Sub-tenant identifier (max 128 chars). See Sub-tenanting.

Response (JSON)

{
  "paragraph_id": "pg_abc123def456",
  "chars_received": 412,
  "claims": [ /* array of Claim objects, each fully verified */ ],
  "counts": {
    "total": 3,
    "confirmed": 2,
    "disputed": 0,
    "inconclusive": 1,
    "skipped": 0
  },
  "processing_time_ms": 9420
}
Retry contract: retry on 5xx and network errors with backoff (recommended: 3 attempts, 0.25s / 0.6s / 1.0s). A 4xx response means the paragraph was rejected and a retry will produce the same error.

POST /api/v1/verify/paragraph/stream

Streaming variant of /verify/paragraph. Same input, but the response is a server-sent-events (SSE) stream so claim cards can render before verdicts arrive.

POST /api/v1/verify/paragraph/stream Stateless single-paragraph verify (SSE)

Request Body

Same as /verify/paragraph.

SSE Events

Events arrive in this order:

EventDescription
paragraph_claimsFires once after extraction. Payload includes the extracted claims with their span/sentence offsets but no verdicts yet. Use to render placeholder cards immediately.
verify_resultFires once per claim as each verification finishes (out of input order — fastest claim arrives first). Payload is a fully-populated Claim object.
doneFires once at the end with paragraph_id, chars_received, total_claims, and processing_time_ms.
errorFired if verification fails for the whole paragraph (rare; per-claim failures arrive as verify_result with verdict: "inconclusive").

Example event payload

event: paragraph_claims
data: {"paragraph_id": "pg_abc123def456", "claims": [{"text": "...", "sentence": "...", "start_offset": 0, "end_offset": 49}]}

event: verify_result
data: {"text": "...", "verdict": "confirmed", "summary": "...", "explanation": "...", "sources": [...], "verified_by": "Sonar"}

event: done
data: {"paragraph_id": "pg_abc123def456", "chars_received": 412, "total_claims": 3, "processing_time_ms": 9420}
Use this when you want progressive UI updates (claim cards appear before verdicts). Use the JSON variant when you only need the final result.

POST /api/v1/extract

Extract individual claims from content using ProRata's extraction engine. No verification is performed — returns claims only. Significantly faster and cheaper than the verify endpoint.

POST /api/v1/extract Extract claims (no verification)

Request Body

ParameterTypeRequiredDescription
content string Required The text to extract claims from. Maximum length depends on plan.
content_type string Required Input format: text, url, youtube, tiktok, pdf, docx, image.
max_claims integer Optional Maximum claims to extract (1–2000). Default: 2000.
context string Optional Additional context to improve extraction accuracy.
skip_table_claims boolean Optional Filter out granular table/statistical claims. Default: true.

Response

{
  "id": "ex_abc123def456",
  "status": "complete",
  "claims": [
    {
      "text": "The Earth is approximately 4.5 billion years old",
      "sentence": "The Earth is approximately 4.5 billion years old.",
      "start_offset": 0,
      "end_offset": 49,
      "filtered": false
    }
  ],
  "claims_count": 1,
  "processing_time_ms": 820,
  "usage": {
    "claims_extracted": 1,
    "claims_filtered": 0,
    "content_length": 49,
    "cost_usd": 0.002
  }
}

POST /api/v1/extract-text

Extract plain text from a file or URL. No claim extraction, no verification, no billing. Returns the extracted text with a human-readable document title. Use this to preview content before running it through the verify endpoint.

POST /api/v1/extract-text Extract text and title (free)

Request Body

ParameterTypeRequiredDescription
content_type string Required Input format: text, url, youtube, tiktok, pdf, docx, image.
content string Conditional Raw text input. Required when content_type is text.
url string Conditional URL to fetch. Required when content_type is url, youtube, or tiktok.
file string Conditional Base64-encoded file content. Required when content_type is pdf, docx, or image.

Response

FieldTypeDescription
id string Unique job identifier (et_...).
status string Always "complete".
input_type string The content_type that was submitted.
text string Extracted plain text. DOCX preserves headings as markdown (#/##), lists, and tables.
title string Human-readable document title. Extracted from file metadata, first heading, or cleaned filename. For YouTube, returns the video title. For URLs, returns the page title.
char_count integer Length of extracted text in characters.
processing_time_ms integer Server-side processing time in milliseconds.

Example

# Extract text from a DOCX file
curl -X POST https://api.factivelabs.com/api/v1/extract-text \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "content_type": "docx",
    "file": "UEsDBBQAAAAIAB..."
  }'
{
  "id": "et_a1b2c3d4e5f6",
  "status": "complete",
  "input_type": "docx",
  "text": "# Introduction\n\nThis thesis examines...",
  "title": "Czech Independence and the Chicago School",
  "char_count": 48210,
  "processing_time_ms": 340
}

Title Extraction

The title field uses a waterfall strategy per content type:

Content TypeTitle Source (in priority order)
docxDocument metadata (dc:title) → first Heading 1 → cleaned filename
pdfPDF metadata title → cleaned filename
urlPage og:title → <title> tag → URL path
youtubeVideo title from YouTube
tiktok"TikTok video"
imageCleaned filename
textFirst non-empty line (truncated to 100 chars)

POST /api/v1/analyze-structure

Analyze a document's section structure and get smart skip recommendations. Zero LLM calls — pure text parsing, returns in milliseconds. Use this to build a section picker UI before running the expensive verify endpoint.

POST /api/v1/analyze-structure Analyze document sections (free)

Request Body

ParameterTypeRequiredDescription
content string Conditional Document text with markdown headings. Required when content_type is text. Max 500K chars.
url string Conditional URL to fetch. Required when content_type is url.
file string Conditional Base64-encoded file. Required when content_type is pdf or docx.
content_type string Optional Input format: text (default), url, pdf, docx.

Response

{
  "sections": [
    {
      "index": 0,
      "name": "Frontmatter",
      "level": 0,
      "start_char": 0,
      "end_char": 412,
      "word_count": 65,
      "recommended": false,
      "skip_reason": "Frontmatter (title, author, abstract)",
      "child_indices": [0]
    },
    {
      "index": 1,
      "name": "Introduction",
      "level": 1,
      "start_char": 412,
      "end_char": 3842,
      "word_count": 580,
      "recommended": true,
      "skip_reason": null,
      "child_indices": [1, 2, 3]
    }
  ],
  "total_sections": 9,
  "total_words": 13420,
  "recommended_sections": 6,
  "recommended_words": 10800,
  "source_hash": "a1b2c3d4e5f6"
}

Pass the section indices to /api/v1/verify via include_sections or exclude_sections to fact-check only the sections you want. The source_hash lets you verify the text hasn't changed between analyze and verify calls.

GET /api/v1/jobs/{job_id}

Check the status of an async verification job. Poll this endpoint after submitting a request with "async": true.

GET /api/v1/jobs/{job_id} Check async job status

Path Parameters

ParameterTypeRequiredDescription
job_id string Required The job ID returned from the verify endpoint when using async mode.

Response (Processing)

{
  "id": "fc_abc123def456",
  "status": "processing",
  "progress": {
    "claims_extracted": 8,
    "claims_verified": 3,
    "claims_total": 12
  }
}

Response (Completed)

{
  "id": "fc_abc123def456",
  "status": "complete",
  "progress": {"claims_extracted": 12, "claims_verified": 12, "claims_total": 12},
  "result": {
    "id": "fc_abc123",
    "status": "complete",
    "title": "Czech Independence and the Chicago School",
    "claims": [...],
    "counts": {"total": 12, "confirmed": 8, "disputed": 2, "inconclusive": 2, "skipped": 0},
    "usage": {"claim_count": 12, "skipped_claims": 0, "cost_usd": 0.12}
  },
  "created_at": "2026-04-11T10:30:00Z",
  "completed_at": "2026-04-11T10:30:45Z"
}

Response (Failed)

{
  "id": "fc_abc123def456",
  "status": "failed",
  "progress": {"claims_extracted": 0, "claims_verified": 0, "claims_total": 0},
  "result": null,
  "created_at": "2026-05-14T10:30:00Z",
  "completed_at": "2026-05-14T10:30:01Z"
}

POST /api/v1/verify/batch

Submit multiple documents for verification in a single request. Each item is processed independently.

POST /api/v1/verify/batch Batch verify multiple documents

Request Body

ParameterTypeRequiredDescription
items array Required Array of verification requests. Each item has content and content_type fields (same schema as the verify endpoint). Max items: Free 5, Pro 50, Enterprise 100.

Request Example

{
  "items": [
    {"content": "The speed of light is 300,000 km/s", "content_type": "text"},
    {"url": "https://example.com/article", "content_type": "url"},
    {"content": "Water boils at 100 degrees Celsius at sea level", "content_type": "text"}
  ]
}

Response

{
  "results": [
    {"id": "ver_001", "status": "completed", "claims": [...], "usage": {...}},
    {"id": "ver_002", "status": "completed", "claims": [...], "usage": {...}},
    {"id": "ver_003", "status": "completed", "claims": [...], "usage": {...}}
  ],
  "total_usage": {
    "claims_extracted": 5,
    "claims_verified": 5,
    "items_processed": 3
  }
}

GET /health

Health check endpoint. No authentication required.

GET /health Health check (no auth)

Response

{"status": "ok", "version": "1.0.0"}

Private Corpus

Customer-private document corpus. Verify against your own documents instead of (or in addition to) the public web. All endpoints require Bearer token authentication. Documents are isolated by API key prefix — one customer cannot read another customer's corpus. If you're sub-tenanting, isolation is further scoped to (api_key, end_user_id) so each of your end-users gets their own corpus under your account.

How to use the corpus from /verify: set "use_private_corpus": true on your POST /api/v1/verify or POST /api/v1/verify/paragraph request. Combine with "corpus_mode" to control the fall-back behavior.
GET /api/v1/private-corpus List all live documents for this customer

Returns the customer's documents (newest first), excluding deleted ones. Each item includes the document ID, filename, content type, byte size, character count, chunk count, ingestion status, and timestamps.

Response

{
  "docs": [
    {
      "doc_id": "doc_abc123",
      "filename": "policies.pdf",
      "content_type": "pdf",
      "size_bytes": 184221,
      "char_count": 41092,
      "chunk_count": 88,
      "status": "ready",
      "uploaded_at": "2026-05-08T14:22:11Z"
    }
  ],
  "count": 1
}
POST /api/v1/private-corpus/upload Ingest 1–10 documents asynchronously

multipart/form-data. Send 1–10 files in the files field. Validation happens synchronously; ingestion (text extraction, chunking, embedding) runs as a background task. Returns 202 Accepted with a list of accepted documents in queued state. Poll GET /api/v1/private-corpus to watch each document move queued → processing → embedding → ready.

Supported file types

pdf, docx, doc, rtf, txt, md, html.

DELETE /api/v1/private-corpus/{doc_id} Hard-delete a single document

Hard-delete a single document. Returns 404 if the document does not exist or belongs to a different customer (these are reported the same way to prevent existence-leak).

GET /api/v1/private-corpus/{doc_id}/chunks List the chunks of a document

Returns all chunks of a document, ordered by chunk_index. Used by the dashboard "View" modal so customers can inspect what the fact-checker actually indexed (helps spot OCR mistakes, missing sections, weird chunk boundaries).

POST /api/v1/private-corpus/clear Wipe every document in the corpus

Wipe every document in the customer's corpus. Requires confirm=true in the form body to prevent accidental wipes. Returns the count of documents removed.

GET, PUT /api/v1/private-corpus/scope Read or edit the auto-generated corpus description

GET returns the current scope description (a one-paragraph summary of what the corpus contains, used to improve corpus-routing accuracy when use_private_corpus: true) along with metadata flags edited_by_user, regenerating, and last_regenerated_at. Returns an empty/idle row for customers who have not yet uploaded any documents.

PUT saves a customer-edited description. Sets edited_by_user=true. The next successful upload or delete will overwrite this edit (per design choice 2026-05-03 — auto-regenerate on doc changes).

POST /api/v1/private-corpus/scope/regenerate Manually trigger scope regeneration

Schedules the regeneration as a background task and returns immediately. The dashboard polls GET /api/v1/private-corpus/scope to watch the status flip from regenerating back to idle.

GET /api/v1/private-corpus/audit Recent corpus audit log entries

Return the most recent audit log rows for this customer, newest first. Default limit 100, max 1,000. Rows include the action, document ID, chunk count for verify events, source IP, and success/failure.

Claim Object

Represents a single extracted and verified claim.

FieldTypeDescription
textstringThe extracted claim text as a self-contained, verifiable statement. Note: ProRata rewrites claims with coreference resolution (e.g. "it" becomes "The Great Wall"), so this may differ from the original text.
sentencestringOriginal sentence from which the claim was extracted.
sentence_textstringFull source sentence text (may differ slightly from sentence in edge cases).
start_offsetinteger|nullCharacter start position of the claim region in the source text.
end_offsetinteger|nullCharacter end position of the claim region in the source text.
sentence_startinteger|nullCharacter start of the full source sentence.
sentence_endinteger|nullCharacter end of the full source sentence.
span_startinteger|nullSub-span start within the sentence (for multi-claim sentences).
span_endinteger|nullSub-span end within the sentence.
claim_typestring"factual" or "subjective".
verdictstring|nullOne of: confirmed, disputed, inconclusive, or null if not yet verified.
summarystringOne-sentence verdict summary.
explanationstringDetailed explanation of why the claim received this verdict.
corrected_textstring|nullSuggested correction for disputed claims (null if confirmed or no correction available).
research_querystringThe verification query sent to the research model.
categoriesarray[string]Claim topic categories (e.g. "science", "politics", "historical").
difficultyintegerHow hard to fact-check (1=trivial, 3=needs sources, 5=highly ambiguous). 0=not classified.
verified_bystringIdentifier for the verification path that produced the verdict. Common values include "exa+haiku" (default web verifier), "corpus+haiku" (private corpus verifier), "gemini" (fallback verifier), "Sonnet" (second-pass on disputed claims), and "Sonar" / "sonar_bo3" (legacy paths). Useful for debugging or for the Gemini-fallback caveat noted in Source Blacklist.
skippedbooleantrue if the claim was auto-skipped (non-verifiable framing such as idioms, meta-framing, or unintelligible text). Skipped claims have verdict: "skipped" and are not billed.
context_flagsobjectFraming flags detected by the context analyzer. Common keys: category (string — e.g. "science", "politics"), needs_context (boolean — true if the claim depends on disambiguation), skip_flag (string — "none", "discourse_claim", "idiom", "unintelligible"). Additional internal flags may also appear depending on the pipeline path.
unclear_reasonstring|nullFor inconclusive verdicts, indicates why: "NO_SOURCES", "INSUFFICIENT", "CONFLICTING", or "SUBJECTIVE". Null for non-inconclusive verdicts.
sourcesarray[Source]Source citations supporting the verdict.

ExtractedClaim Object

Represents a single extracted claim (no verification data).

FieldTypeDescription
textstringThe extracted claim text, self-contained and verifiable.
sentencestringOriginal sentence from which the claim was extracted.
start_offsetintegerCharacter start position in the source text (nullable).
end_offsetintegerCharacter end position in the source text (nullable).
filteredbooleanTrue if the claim was flagged as granular table data by the filter.

Highlight Object

Represents a pre-resolved region in the source text where a verified claim was found. Returned in the highlights SSE event after all claims are verified. Overlaps are already resolved (verdict priority: disputed > inconclusive > confirmed). See Claim Positioning and the Integration Guide in the documentation.

FieldTypeDescription
startintegerCharacter start offset in the source text.
endintegerCharacter end offset in the source text.
verdictstringOne of: confirmed, disputed, inconclusive.
textstringRecommended primary locator. The exact words to highlight. Find this in your rendered content with indexOf() — works regardless of rendering differences. See the Integration Guide.
answer_spanstringThe full original sentence containing the claim. Fallback search key when text can't be found in the local copy.
claim_indexintegerIndex of the primary claim this region corresponds to.
claim_indicesarray[integer]All claim indices covered by this region (when multiple claims share the same span).
tooltipstringShort display text suitable for hover tooltips.

Highlights Envelope

The highlights SSE event wraps the array of Highlight objects with two fields for divergence detection:

FieldTypeDescription
highlightsarray[Highlight]Sorted by start position. Non-overlapping.
source_lengthintegerCharacter length of the source text at computation time. Compare against your local copy to detect truncation.
source_hashstringMD5 fingerprint (first 12 hex chars) of the source text. If your local hash differs, fall back to text or answer_span for positioning.

Source Object

Source metadata for a citation. All fields default to empty string (never null).

FieldTypeDescription
urlstringSource URL.
titlestringPage title.
domainstringDomain name (e.g. "reuters.com"). May be empty on responses from the default exa+haiku verifier path — parse the hostname from url as a fallback.
publisherstringPublisher name derived from domain (e.g. "Reuters"). May be empty on the exa+haiku path — same fallback as domain.
snippetstringRelevant excerpt from the source.
datestringPublication date (if available from search results).

Usage Object

Returned by the /api/v1/verify endpoint.

FieldTypeDescription
claim_countintegerTotal claims extracted and verified.
skipped_claimsintegerClaims auto-skipped by the table filter (not billed).
cost_usdfloatCost for this request in USD.

Extract Usage Object

Returned by the /api/v1/extract endpoint.

FieldTypeDescription
claims_extractedintegerTotal claims extracted.
claims_filteredintegerClaims filtered out by the table filter.
content_lengthintegerCharacter length of the input content.
cost_usdfloatCost for this request in USD.

Error Object

FieldTypeDescription
error.codestringMachine-readable error code (e.g., rate_limit_exceeded, invalid_api_key).
error.messagestringHuman-readable error description.
error.retry_afterintegerSeconds to wait before retrying (for rate limit errors).