Getting Started

Question in. Fair answer + certificate out. Three lines of code.

Haven Shield API

One API call. Your user's question goes in. A fair, knowledge-backed response comes out — with a score, a knowledge depth badge, and a compliance certificate.

Quick Start — Python

import requests

response = requests.post("https://haven-voice-ai.amitacompany.workers.dev/api/v1/generate",
    json={"query": "What are the health benefits of Islamic prayer?"}).json()

print(response["response"])                          # Fair, cited answer
print(response["score"]["knowledge_depth"]["label"]) # "Deep knowledge"
print(response["certificate"]["id"])                 # cert_xxxxx

Quick Start — JavaScript

const res = await fetch("https://haven-voice-ai.amitacompany.workers.dev/api/v1/generate", {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({ query: "What are the health benefits of Islamic prayer?" })
});
const data = await res.json();

console.log(data.response);                          // Fair, cited answer
console.log(data.score.knowledge_depth.label);       // "Deep knowledge"
console.log(data.certificate.id);                    // cert_xxxxx

Quick Start — cURL

curl -X POST https://haven-voice-ai.amitacompany.workers.dev/api/v1/generate \
  -H "Content-Type: application/json" \
  -d '{"query": "What are the health benefits of Islamic prayer?"}'

Response Format

{
  "response": "The five daily Islamic prayers (salat) have been...",
  "score": {
    "z_task": 0.89,
    "fair": true,
    "knowledge_depth": {
      "level": "deep",
      "color": "green",
      "sources": 15
    },
    "facts_used": 10,
    "facts_backed": 8
  },
  "certificate": {
    "id": "cert_abc123",
    "fair": true,
    "compliance": {
      "ferpa": "PASS",
      "coppa": "PASS",
      "gdpr": "PASS"
    }
  }
}

Parameters

Parameter	Type	Default	Description
`query`	string	required	The question to answer
`d`	number	0.85	Fairness threshold (0.0–1.0). Below this = response flagged as unfair
`max_tokens`	number	500	Maximum response length

Knowledge Depth Badge

Every response includes a knowledge depth indicator — the lock icon for AI:

Deep knowledge — 15+ verified sources backing this response
Moderate knowledge — 5-14 verified sources
Thin knowledge — fewer than 5 verified sources

Display the badge in your UI so users can see how much the AI knows about each topic.

Live Demo

See it in action: Haven Shield Live Demo — pick a topic, one click, see the gap.

Paragon Fairness — Docker (On-Premise)

For teams that need to train fair models on their own infrastructure. Data never leaves your machine.

Prerequisites

Docker Desktop installed and running
A GitHub access token (provided by Paragon Biosignals)
Your data as a CSV file

Step 1: Authenticate

echo YOUR_ACCESS_TOKEN | docker login ghcr.io -u YOUR_GITHUB_USERNAME --password-stdin

One-time setup. Docker remembers the credentials.

Step 2: Pull the Image

docker pull ghcr.io/paragon-dao/fairness:latest

Step 3: Run the Demo

docker run --rm ghcr.io/paragon-dao/fairness:latest demo

Generates synthetic biased data, trains fair models, outputs certificates. About 10 seconds.

Step 4: Train on Your Data

docker run --rm \
  -v /path/to/your/data:/data \
  -v /path/to/output:/output \
  ghcr.io/paragon-dao/fairness:latest train \
    --input /data/your_data.csv \
    --sensitive-column ancestry \
    --target-column disease_risk \
    --d-sweep \
    --output /output

What `--d-sweep` Does

Trains 5 models at d=8, 16, 32, 64, 128. Lower d = fairer. Higher d = more accurate. You pick the trade-off.

Step 5: Verify a Certificate

docker run --rm \
  -v /path/to/output:/output \
  ghcr.io/paragon-dao/fairness:latest verify \
    --certificate /output/fairness_certificate_d32.json

Step 6: Upgrade

Feature	Free	Pro ($1,500/mo)	Enterprise (Contact Sales)	Regulated (Contact Sales)
Samples	1,000	Unlimited	Unlimited	Unlimited
d values	32, 64	Full sweep	Full sweep	Full sweep
Certificate	Watermarked	Signed	Signed	Signed
Audit trail	-	-	Yes	Yes
Compliance	-	-	-	EU AI Act / FDA

Data Privacy

Your data never leaves your machine. The Docker runs locally.
No telemetry, no phone-home, no data collection.
Only the Fairness Certificate (metrics JSON) can optionally be submitted to paragondao.org/verify.
The Docker image contains compiled binaries only — no readable source code.

Support

General: [email protected]
Pro/Enterprise: [email protected]

Patent Notice

Protected by US provisional patents. Evaluation use permitted. Commercial use requires a license from Paragon Biosignals.