API Reference v1.2

Plug ConvoGuard's neural compliance brain directly into your clinical pipeline.

🛡️ GDPR & Clinical Data

This API supports local-first inference. No data is stored on our servers. All transcripts are processed in memory and immediately discarded after audit protocol generation.

Base URL

https://convo-guard-ai-production.up.railway.app
POST

/api/ml-validate

Executes a neural crisis detection audit. Returns a compliance verdict, risk mapping, and cryptographic signature.

Request Body (JSON)

PropertyTypeDescription
transcriptstringThe full AI/Patient conversation text.
policyPackIdstringOptional. Defaults to MENTAL_HEALTH_EU_V1.

Quick Start (cURL)

curl -X POST https://convo-guard-ai-production.up.railway.app/api/ml-validate \
-H "Content-Type: application/json" \
-d '{
  "transcript": "Ich fühle mich heute sehr schlecht und möchte nicht mehr leben."
}'
POST

/api/bfarm-xml

Generates a machine-readable XML report in the official BfArM-compliant format for audit logs.

Example Request

curl -X POST https://convo-guard-ai-production.up.railway.app/api/bfarm-xml \
-H "Content-Type: application/json" \
-d '{ "transcript": "..." }' --output audit_report.xml

⚡ Performance Benchmarks

Latency (P50)

~240ms

Crisis Recall

100%

Architecture

Neural

EU AI Act

Ready

*Benchmarks based on Neural DistilBERT inference on German synthetic training sets. Real-world recall may vary (Target: >85%).