Developers

Evidrai runs programmatically between your AI system and the action it triggers.

The same evidence engine behind Verify is available as a REST API. Send a claim or an AI-generated output; get back a structured, inspectable, machine-readable assessment — or a pass / review / block decision through Gateway Mode.

Evidrai Enterprise Request API access

Authentication

API keys with explicit scopes.

Authenticate with an API key in the X-Evidrai-Api-Key header. Keys carry explicit scopes — assessments:write, reports:read, and gateway:write — so access is granted deliberately. Gateway Mode requires a key that carries gateway:write; session and anonymous callers are rejected.

Endpoints

Core verification surface.

POST/v1/gateway/verify

Verify AI-generated output against evidence and your policy. Returns a pass / review / block decision with the triggered rules and full evidence trail. Requires an API key with the gateway:write scope.

POST/assessments/deep

Retrieval-backed evidence assessment of a single claim. Returns the verdict, confidence, source scoring, contradiction handling, caveats, and inspectable reasoning.

POST/assessments/fast

Lower-latency single-pass assessment for lightweight checks.

GET/reports/{id}

Retrieve a previously produced assessment by its persistent verification ID.

Gateway request

One synchronous call.

Send the AI output, the proposed downstream action, the domain, and a consequence level. Supply your own cited_sources and they are folded into the evidence set, scored by the same engine as the open web.

curl -sS https://api.evidrai.com/v1/gateway/verify \
  -H 'Content-Type: application/json' \
  -H 'X-Evidrai-Api-Key: evd_live_...' \
  -d '{
    "workflow_id": "investment-briefing-prod",
    "request_id": "req-123",
    "ai_output": "The claim your AI system produced.",
    "proposed_action": { "type": "publish_briefing", "description": "Publish briefing" },
    "domain": "finance",
    "consequence_level": "high",
    "policy_id": "finance_default_v1",
    "cited_sources": []
  }'

Gateway response

A structured decision.

{
  "schema_version": "gateway_decision.v1",
  "gateway_id": "gw_...",
  "decision": "review",
  "decision_reason": "High-consequence output has a material claim that is only partly supported.",
  "recommended_next_step": "route_to_human_review",
  "policy_id": "finance_default_v1",
  "assessment_id": "uuid",
  "triggered_rules": [ { "rule_id": "material_claim_unverified", "severity": "review" } ],
  "blocking_claims": [],
  "review_claims": ["sc_1"]
}

Assessment response

Machine-readable and inspectable.

Every assessment is a typed JSON contract (schema_version: assessment_response.v1) with a persistent ID, so results are auditable and re-retrievable.

assessment_id — persistent verification ID
verdict — label, confidence, confidence_score, key_caveat, evidence_strength_score
claim_breakdown — decomposed subclaims with supporting / contradicting source IDs
sources — provenance, source type, and per-source scoring factors
evidence_map — sources grouped by role (supports, contradicts, context)
reasoning — inspectable rationale, rule engine, amplification warnings
telemetry — model, tokens, search calls, elapsed_ms, estimated_cost_usd

Access

API access is provisioned for Enterprise pilots.

Programmatic and Gateway access are granted per engagement. Tell us about your workflow and we will set up a scoped key.

Request API access Learn about Enterprise