Zero Point Logic • AI Bias Filter

ZPL AI Filter
Every AI response,
bias-certified

Pass any AI provider through ZPL's bias-neutrality engine. Your messages, your API key — ZPL analyzes every response for ideological lean and optionally rebalances it toward true equilibrium.

Try it live View API docs
3
AI Providers
AIN
Score 0–1.0
0
Keys Stored
HTTPS
End-to-End
Process

How ZPL AI Filter works

Three steps from raw AI output to bias-certified response.

01
🔑

Your API key — your cost

Send your own OpenAI, Anthropic, or Groq API key with every request (BYOK). ZPL never stores it — it's used once, in memory, and discarded immediately after the call.

02
🤖

AI generates its response

ZPL forwards your messages to the provider. The AI generates its response normally. ZPL acts as a transparent middleware — your messages are never modified before delivery.

03
⚖️

ZPL scores & filters

The response is tokenized and analyzed for sentiment imbalance. An AIN Score (0.0 = biased, 1.0 = neutral) is computed using ZPL's equilibrium algorithm. In rebalance or strict mode, biased responses are automatically retried with a neutralizing system prompt.

0.0 – 0.4 · Biased
Response heavily favors one polarity. Rebalance mode triggers a retry.
0.4 – 0.7 · Moderate
Some imbalance detected, but within tolerable range for many use cases.
0.7 – 1.0 · ZPL Certified ✓
Response is at or near equilibrium. ZPL certified neutral badge applied.
Interactive Tester

Try the ZPL AI Proxy

Select a provider, paste your API key, write a message, and see bias analysis in real time.

Analyze: Score the response without modifying it. See the raw AIN bias score.

AI Response

ZPL Fairness Report

0.0 0 1
—
AIN Score
neutral
Applications

Built for high-stakes domains

Wherever AI bias can cause real harm, ZPL filters it out.

👤

HR & Recruitment

Ensure AI screening tools don't systematically favor or reject candidates based on biased language in job descriptions or evaluations.

📰

Content Moderation

Verify that AI content classifiers apply standards evenly across topics, communities, and ideological perspectives.

💹

Financial Advice

Detect when AI financial guidance leans toward overly optimistic or pessimistic framing that could influence user decisions.

🩺

Medical Q&A

Flag responses that over-reassure or catastrophize. Balanced medical information respects patient autonomy.

⚖️

Legal Research

Ensure AI legal summaries present both sides of case law and argument without nudging toward a predetermined conclusion.

🎓

Education Tools

Keep AI tutors and essay assistants pedagogically neutral, presenting multiple viewpoints rather than steering students.

Security

Your API key stays yours

We built the proxy around one non-negotiable principle: ZPL never stores your AI API keys.

🔒

HTTPS Only

All traffic between your browser, ZPL's backend, and the AI provider is encrypted via TLS. No plain-text connections accepted.

⚡

Used Once, Gone

Your API key lives in memory for the duration of one HTTP request — typically under 5 seconds. It is never written to disk, database, logs, or cache.

🚫

Never Logged

ZPL's usage logs track only: timestamp, provider name, status code, and response time. API key values are explicitly excluded from all log entries.

🛡️

JWT Auth Required

The proxy endpoint requires a valid ZPL account JWT. Rate limits prevent abuse. Enterprise users get dedicated throughput quotas.

Security statement: Your AI API key is transmitted over HTTPS directly to ZPL's backend and immediately used to make a single request to the AI provider. It is never saved to a database, written to disk, included in logs, or retained in any form after the request completes. ZPL's source code is available for audit upon enterprise request.
Plans

AI Proxy call limits

Proxy calls are counted separately from ZPL compute calls. Upgrade anytime.

Free
$0
forever
0 calls
AI proxy calls / month
  • AI Proxy access
  • ZPL compute (N≤9)
  • ZPL API access
  • API keys
Basic
$12
/ month
100 calls
AI proxy calls / month
  • AI Proxy (100/mo)
  • ZPL compute (N≤16)
  • 3 API keys
  • Usage dashboard
Pro
$49
/ month
1,000 calls
AI proxy calls / month
  • AI Proxy (1,000/mo)
  • ZPL compute (N≤64)
  • 5 API keys
  • Strict mode & sweep
  • Priority support
Enterprise
Custom
contact us
Unlimited
AI proxy calls / month
  • Unlimited AI Proxy
  • All N sizes
  • 20 API keys
  • SLA & audit logs
  • Dedicated support
  • Source code audit
Get started free
Developer API

Integrate in minutes

Use the ZPL proxy from any language. Authenticate with your ZPL JWT token.

import requests

# Your ZPL JWT token (from /auth/login)
ZPL_TOKEN = "eyJhbGciOiJIUzI1NiIs..."

# Your own Groq key (free at console.groq.com)
MY_GROQ_KEY = "gsk_your_groq_key_here"

response = requests.post(
    "https://zpl-backend.onrender.com/ai/proxy",
    headers={"Authorization": f"Bearer {ZPL_TOKEN}"},
    json={
        "provider": "groq",
        "model": "llama3-8b-8192",
        "messages": [
            {"role": "user", "content": "What are the pros and cons of nuclear energy?"}
        ],
        "user_api_key": MY_GROQ_KEY,
        "zpl_options": {
            "filter_mode": "rebalance"  # analyze | rebalance | strict
        }
    }
)

data = response.json()
print(data["response"])
print(f"AIN Score: {data['zpl_filter']['ain_score']}")
print(f"ZPL Certified: {data['zpl_filter']['zpl_certified']}")
print(f"Bias direction: {data['zpl_filter']['bias_direction']}")
// Using the built-in ZplApi client (api.js)
// Make sure user is logged in (ZplApi.token is set)

const result = await ZplApi.aiProxy(
  'groq',                     // provider
  'llama3-8b-8192',          // model
  [{ role: 'user', content: 'Pros and cons of nuclear energy?' }],
  'gsk_your_groq_key',       // your API key — never stored
  'rebalance'                // filter mode
);

console.log(result.response);
console.log('AIN Score:', result.zpl_filter.ain_score);
console.log('Certified:', result.zpl_filter.zpl_certified);

// Or raw fetch:
const res = await fetch('https://zpl-backend.onrender.com/ai/proxy', {
  method: 'POST',
  headers: {
    'Authorization': `Bearer ${zplToken}`,
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    provider: 'openai',
    model: 'gpt-4o',
    messages: [{ role: 'user', content: '...' }],
    user_api_key: 'sk-...',
    zpl_options: { filter_mode: 'strict' }
  })
});
const data = await res.json();
# Analyze mode — just score, no modification
curl -X POST https://zpl-backend.onrender.com/ai/proxy \
  -H "Authorization: Bearer YOUR_ZPL_JWT" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "groq",
    "model": "llama3-8b-8192",
    "messages": [
      {"role": "user", "content": "What are the pros and cons of nuclear energy?"}
    ],
    "user_api_key": "gsk_YOUR_GROQ_KEY",
    "zpl_options": {"filter_mode": "analyze"}
  }'

# Example response:
# {
#   "response": "Nuclear energy has several advantages...",
#   "provider": "groq",
#   "model": "llama3-8b-8192",
#   "zpl_filter": {
#     "mode": "analyze",
#     "ain_score": 0.847,
#     "bias_direction": "neutral",
#     "attempts": 1,
#     "was_rebalanced": false,
#     "zpl_certified": true
#   },
#   "usage": {"proxy_calls_used": 12, "proxy_calls_limit": 100}
# }

# Standalone text analysis (no AI call needed):
curl -X POST https://zpl-backend.onrender.com/ai/analyze \
  -H "Authorization: Bearer YOUR_ZPL_JWT" \
  -H "Content-Type: application/json" \
  -d '{"text": "This is definitely the best solution available. You should always choose it."}'

Response Schema

{
  "response":  // string — the AI's full response text
  "provider":  // "openai" | "anthropic" | "groq"
  "model":     // model used (echoed back)
  "zpl_filter": {
    "mode":              // "analyze" | "rebalance" | "strict"
    "ain_score":         // float 0.0–1.0 (higher = more neutral)
    "bias_direction":    // "neutral" | "positive_leaning" | "negative_leaning"
    "attempts":          // 1–3 (how many AI calls were made)
    "was_rebalanced":    // bool — true if ZPL retried with neutral prompt
    "pos_ratio":         // fraction of sentiment words that are positive
    "neg_ratio":         // fraction of sentiment words that are negative
    "word_count":        // total words in response
    "sentiment_words_found": // count of detected sentiment vocabulary
    "zpl_certified":    // bool — true if ain_score >= 0.7
  },
  "usage": {
    "proxy_calls_used":  // your running total this month
    "proxy_calls_limit": // your plan's monthly limit ("unlimited" for enterprise)
  }
}