Every client.fetch() routes
through HIVECOMPUTE.
Every reply ships with a receipt.

A drop-in funnel. Not an abstraction layer.

hive-rosetta is a thin wrapper. It takes three lines of code to replace your existing OpenAI SDK calls. The API shape matches the OpenAI Python, Node, and Go SDKs exactly. What's different is where the call goes and what comes back. Every response includes a X-Spectral-ZK-Receipt header holding the Ed25519-signed audit record.

Inference endpoint: srotzin--hivecompute-hivecompute-server.modal.run/v1/compute/chat/completions

Three lines. Identical API shape. Every call signed.

The hive-rosetta client mirrors the OpenAI SDK interface exactly. If your codebase uses openai.chat.completions.create(), the migration is a one-line import change.

Python

# Before: standard OpenAI client
# from openai import OpenAI
# client = OpenAI(api_key=...)

from hive_rosetta import Client   # drop-in replacement

client = Client(api_key="your-hive-key")

response = client.chat.completions.create(
    model="hivecompute-g2g7",
    messages=[{"role": "user", "content": "Analyze this contract."}]
)

# response.choices[0].message.content: same shape as the OpenAI response
# response.spectral_receipt: Spectral-ZK receipt dict, attached automatically
receipt = response.spectral_receipt
print(receipt["id"])          # spectral_zkr_01HZ9XK...
print(receipt["hash"])        # b3:b9f0e147...
print(receipt["signatures"])  # Ed25519 sig from registrar DID

Node.js / TypeScript

// Before: standard OpenAI client
// import OpenAI from 'openai';
// const client = new OpenAI({ apiKey: ... });

import { HiveClient } from 'hive-rosetta';  // drop-in replacement

const client = new HiveClient({ apiKey: 'your-hive-key' });

const response = await client.chat.completions.create({
  model: 'hivecompute-g2g7',
  messages: [{ role: 'user', content: 'Summarize this document.' }]
});

// response.choices[0].message.content: same shape as the OpenAI response
// response.spectralReceipt: Spectral-ZK receipt object, attached automatically
const receipt = response.spectralReceipt;
console.log(receipt.id);          // spectral_zkr_01HZ9XK...
console.log(receipt.hash);        // b3:b9f0e147...
console.log(receipt.signatures);  // Ed25519 sig from registrar DID

Skipping the verified path isn't just riskier. It puts you in a worse tier everywhere.

Routing outside hive-rosetta doesn't just mean you're missing a receipt. It means you're stuck in a lower tier on every single thing the clearinghouse tracks. The table below shows exactly what changes between a Rosetta-verified agent and one running an unverified path.

That 0.25% fee difference on x402 settlements adds up fast at volume. At 100,000 transactions a month averaging $100 each, Rosetta saves you $25,000 a month in settlement fees alone, before you even count compliance costs.

Here's what the embedded receipt looks like, and why the EU AI Act requires it.

On every hive-rosetta call, HIVECOMPUTE attaches a signed receipt to the response. It's the Spectral-ZK record, the same one described on the HIVEAUDIT mandate page. EU AI Act Article 12 says high-risk AI systems have to "technically allow for the automatic recording of events (logs) over the lifetime of the system." The embedded receipt is that record.

{
  "id":        "spectral_zkr_01HZ9XKDMN2XV9C1B7E8D4F3AU",
  "type":      "SpectralZKReceipt",
  "version":   "hiveaudit-v1",
  "agent_did": "did:hive:agent-us-inference-a4f91c",
  "model":     "hivecompute-g2g7",
  "issued_at": 1746294118472,
  "hash":      "b3:b9f0e147ad53c61f8e9c2b7d4a5163e4f7c9b1d8e3a6f25c0b9d4e7a138f6c2b",
  "canon":     "jcs-rfc8785",
  "hash_alg":  "blake3-256",
  "signatures":[{"signer":"did:hive:registrar-us-hive-001","alg":"ed25519","sig":"5a91e8d7..."}],
  "merkle_root":"7d4a5163e4f7c9b1d8e3a6f25c0b9d4e7a138f6c2b5a91e8d7f3c0a9b4e62d18"
}

The receipt is also accessible via response.spectral_receipt (Python) or response.spectralReceipt (Node). The full receipt JSON is appended to the HIVEAUDIT Reckoning chain immediately on issuance.

Cite: EU AI Act Art. 12: artificialintelligenceact.eu/article/12

Route through Rosetta and the HIVEAUDIT meter turns on by itself.

There's no separate setup step. The moment your first hive-rosetta call hits HIVECOMPUTE, the HIVEAUDIT receipt meter starts running. You're billed per call, not per seat.

See the full pricing breakdown on the HIVEAUDIT mandate page. The Pilot tier (10K receipts/month) is available at $0/month for evaluation.

You get it set up for your environment, through a private briefing.

hive-rosetta runs against a private package index scoped to your environment, jurisdiction, and DID. Set up a private briefing, and you'll get your install commands, fee tier, and SHOD-gate limits as part of that briefing. We plan to make it generally available on PyPI and npm down the road.

# Provisioned per-environment via briefing
pip install hive-rosetta \
  --extra-index-url https://pypi.thehiveryiq.com/simple

# Provisioned per-environment via briefing
npm install hive-rosetta \
  --registry https://npm.thehiveryiq.com

Design your private briefing for hive-rosetta.
Your first signed receipt within four minutes of provisioning.

Every call, every receipt, and every agent DID gets logged and signed. The audit chain starts the moment your first call hits HIVECOMPUTE.