import OpenAI from 'openai';
import { auto } from '@openrng/auto';

// Wrap your existing client — one line
const client = auto(new OpenAI());

// Use exactly as before — VEOs are emitted automatically
const response = await client.chat.completions.create({
  model: 'gpt-4o',
  messages: [{ role: 'user', content: 'What is the capital of France?' }],
});
// response is unchanged — same OpenAI response object
// but a VEO-2 record was created in the background

With Anthropic

import Anthropic from '@anthropic-ai/sdk';
import { auto } from '@openrng/auto';

const client = auto(new Anthropic());

const message = await client.messages.create({
  model: 'claude-4-sonnet',
  messages: [{ role: 'user', content: 'Hello' }],
});

With Signing

import { auto } from '@openrng/auto';
import { generateSigningKeys } from '@openrng/core';

const keys = generateSigningKeys();
const client = auto(new OpenAI(), {
  privateKey: keys.privateKey,
  provider: 'my-service',
});
// Every VEO is now Ed25519-signed and independently verifiable

Accessing VEOs

import { auto, MemoryStore } from '@openrng/auto';

const store = new MemoryStore();
const client = auto(new OpenAI(), { store });

await client.chat.completions.create({ ... });

// Get all recorded VEOs
const veos = store.list();
console.log(veos[0].execution?.model_id); // 'gpt-4o'
console.log(veos[0].execution?.cost);     // { total_tokens: 150 }

Callback on Every VEO

const client = auto(new OpenAI(), {
  onVEO: (veo) => {
    console.log('AI call recorded:', veo.object_id);
    // Send to your observability pipeline, database, etc.
  },
});

What Gets Captured

Every instrumented call produces a VEO with:

Field	Source
`execution.prompt_hash`	SHA-256 of the prompt/messages
`execution.output_hash`	SHA-256 of the response
`execution.model_id`	Model from the request
`execution.latency_ms`	Measured response time
`execution.cost`	Token usage from response
`execution.tool_calls`	Tool/function calls if any
`confidence.score`	700 (success) or 100 (error)
`lifecycle.state`	'created' or 'signed'

Instrumented Methods

SDK	Methods
OpenAI	`chat.completions.create`, `completions.create`, `embeddings.create`, `images.generate`
Anthropic	`messages.create`
Generic	`create`, `generate`, `complete`

Safety Guarantees

Never breaks your API calls — VEO emission errors are caught and logged, never thrown
No SDK dependency — works with any client object via ES Proxy
Zero performance impact — VEO creation happens after the API call returns