# VeriAI — The AI Output Integrity Layer

> VeriAI scores, hashes, and permanently anchors any AI-generated output. One API call. No auth. Free to start.

## What VeriAI Does

VeriAI is an AI output integrity platform. When you call any AI model (OpenAI, Anthropic, Google, xAI, Mistral, etc.), you can post the prompt and output to VeriAI to receive:

- A **trust score** from 0 to 100 based on 5 heuristic checks: prompt-output coherence, entropy analysis, semantic consistency, confidence signals, and safety screening
- A **SHA-256 proof hash** — a cryptographic fingerprint of the exact content
- A **permanent, shareable proof URL** (e.g. https://veriai.site/proof/[uuid]) that anyone can verify
- A **public ledger entry** that cannot be altered retroactively
- An **embeddable HTML badge** showing the trust score

## Who Uses VeriAI

- **Developers** building AI apps who want to log, audit, and surface trust scores to users
- **AI agents and autonomous systems** that want to self-verify their outputs
- **Researchers and students** who need a citable, permanent URL for AI-generated content
- **Content platforms** that want to badge AI-assisted content with verifiable proof
- **Enterprise teams** who need audit trails of every AI output their systems produce

## API Integration

VeriAI is designed as a "side call" — you don't change your AI integration at all. After receiving your AI model's response, make one additional call:

```
POST https://veriai.site/api/verifications
Content-Type: application/json

{
  "provider": "OpenAI",
  "model": "gpt-4o",
  "prompt": "your original prompt here",
  "output": "the AI response here"
}
```

Response (within ~50ms):
```json
{
  "id": "uuid",
  "trustScore": 84,
  "status": "verified",
  "proofHash": "sha256...",
  "shareableLink": "https://veriai.site/proof/uuid",
  "badgeHtml": "<a href=\"...\">...</a>",
  "checks": [...]
}
```

No API key. No npm install. No account. Works from any language.

## MCP Server (for Claude, Cursor, Windsurf, and AI agents)

VeriAI is available as an MCP (Model Context Protocol) server:

```json
{
  "mcpServers": {
    "veriai": {
      "type": "streamable-http",
      "url": "https://veriai.site/api/mcp"
    }
  }
}
```

Add this to your claude_desktop_config.json or agent config. Claude can then call `verify_output` directly.

## Pricing

- **Free**: 5 verifications per day. No account needed. Full API and MCP access.
- **Pro**: $29 one-time. Unlimited verifications forever. Instant activation. No subscription.

## Key URLs

- Homepage: https://veriai.site
- API endpoint: https://veriai.site/api/verifications
- MCP endpoint: https://veriai.site/api/mcp
- Public ledger: https://veriai.site/ledger
- OpenAPI spec: https://veriai.site/openapi.json
- AI plugin manifest: https://veriai.site/.well-known/ai-plugin.json
- Developer docs: https://veriai.site/developers
- Upgrade: https://veriai.site/upgrade

## Technical Details

- Trust scores use a 5-checkpoint heuristic (not blockchain)
- SHA-256 hash computed server-side over canonicalized content
- Records stored in PostgreSQL with UUID primary keys
- Response time: ~50ms average
- Uptime: monitored via /api/healthz
- CORS: open for all origins (API is public)