Introduction and API model
The vario Scoring API is a stateless REST API that analyzes text for signals of irregular conduct in corporate communications. For each request, it returns a risk score along with structured regulatory context: key phrases, applicable regulatory frameworks, and an action recommendation.
Stateless model
vario does not store or retain the analyzed text. Each call to POST /v1/analyze is independent: the text comes in, is analyzed, and the result is returned in the response. Nothing persists in vario's database.
This design is deliberate. Forensic and e-discovery buyers frequently work with privileged client communications — communications that cannot be held in third-party systems. The stateless model of vario ensures that restriction is not a technical obstacle.
client
│
│ POST /v1/analyze
│ { text, playbook, language, mode }
▼
┌──────────────────────────────────────┐
│ stateless analysis │
│ (text is not stored in DB) │
└──────────────────────────────────────┘
│
│ 200 OK
│ { combined_score, should_alert,
│ severity, reasoning,
│ key_phrases, regulatory_frameworks,
│ recommended_action, ... }
▼
client
Primary use case
The Scoring API is designed for forensic and e-discovery firms that need to classify large volumes of corporate communications to identify which ones require manual review.
A typical scenario: a firm has 80,000 documents in an investigation corpus. Sending all of them to human review is not feasible in terms of time or cost. With the API, the team can process the entire corpus, filter documents with should_alert: true, and focus review on those the system flagged as high-risk.
The API output — the score, key phrases, regulatory framework, and recommendation — is structured to map directly to the workflows of e-discovery platforms such as Relativity or Nuix.
What the API analyzes
The API detects signals of nine types of irregular conduct in corporate communications:
| Conduct | playbook field value | Languages |
|---|---|---|
| Price fixing | price_fixing | es, en |
| Bid rigging | bid_rigging | es, en |
| Bribery and anti-corruption | bribery | es, en |
| Market abuse | market_manipulation | es, en |
| Information exchange | information_exchange | es, en |
| Group boycott | group_boycott | es, en |
| Undisclosed conflict of interest | conflict_of_interest | es, en |
| Insider trading | insider_trading | es, en |
| Internal fraud and embezzlement | internal_fraud | es, en |
These conducts correspond to regulatory categories covered by frameworks such as CADE, COFECE, FNE, CNDC, FCPA, Ley 27.401, Lei Anticorrupção, MAR, and CMF, among others. The API returns the applicable frameworks in the regulatory_frameworks field for each request's analysis.
To query the current catalog of available playbooks, use GET /v1/playbooks.
Analysis modes
Each call to POST /v1/analyze accepts a mode parameter that controls the depth of the analysis and the credit cost:
| Mode | Description | Credits per call | Typical latency |
|---|---|---|---|
screening | L1 layer only (keyword + embedding, no LLM). Mass discard of irrelevant documents. | 1 credit | ~50 ms |
fast | L1 + L2 (light LLM). Classification with moderate precision. | 2 credits | ~200 ms |
full | L1 + L2 + L3 (deep LLM). Analysis with full reasoning and higher precision. | 3 credits | ~2–8 s |
The default mode is "full". For large-volume triage, the recommended practice is a three-stage funnel: "screening" to discard the bulk of the corpus, "fast" on documents that passed the L1 threshold, and "full" for those requiring deep analysis. See Credits & modes for the detailed pattern.
Detection system accuracy
The vario detection system offers validated high precision, with specific coverage for corporate communications in Spanish and Portuguese within a Latin American context. The playbooks account for terminology and language patterns characteristic of cases investigated by regulators such as CADE, COFECE, and FNE.
Get started
See the Quickstart to register, obtain your API key, and send your first request in under five minutes.