HiveBound · Patent Pending

Right now, your AI calls are just questions.
Ours come back signed.

HiveBound wraps every AI call with a declared identity, a scoped intent, and bonded money, all checked before the model runs. Nine checks. Ed25519 signed. HAHS receipt. Settled with x402 on Base.

Who's asking is declared. What they can do is scoped. Money is on the line. Anyone can check the receipt on their own computer, forever. It's not a log. It's a real record.
Preliminary Ed25519 Signed x402 · Base · USDC Patent Pending

Nobody knows who's asking.
That's the problem.

Right now, every AI call is a question with nothing riding on it. A prompt comes in, the model runs, the answer floats out. Nobody signed for it. Nobody's on the hook. There's no proof.

Standard AI calls today

Anonymous question

No declared identity. No approved scope. No money backing it up. So if someone disputes the output later, in an audit, a lawsuit, or a regulatory review, there's nothing to show. Just a log file. Logs can be edited. And logs don't travel with the answer they describe.

A HiveBound call

Signed contract

Every call carries a signed package. It says who asked, what they were allowed to do, and what money they put on the line, all signed before the model even runs. HAHS closes out the receipt. x402 settles the payment on Base. Anyone can check that record on their own computer, forever.

Six pieces.
Wrapped into one package.

HiveBound isn't one single feature. It's six trust pieces working together, and each one covers a different step of the call.

Soul
Identity
AgentPassport
An Ed25519 public key, an agent name, a tenant, and the scopes it's declared. Signed and given a time limit before the call happens.
Mind
Intent
BondedIntent
What this agent is actually allowed to do with the answer: the category, the target model, the provider, the token budget, the scope. Once it's issued, it can't be changed.
Power
Execution
Your inference provider
Works with any OpenAI-compatible endpoint: Fireworks, Together, OpenAI. The HiveBoundAdapter wraps around it. You keep your same provider.
Reality
Proof
SHOD / HAHS Receipt
An Ed25519-signed receipt, locked to its data set, that shows if anyone tampered with it. You can check it on your own computer, forever. DR5 catches anyone who swaps the underlying data.
Space
Settlement
x402 · USDC · Base
Runs on x402 payment rails on Base (chain ID 8453). USDC only. A 5bps fee goes to the treasury on each call. The money that was bonded gets settled on-chain.
Time
Provenance
Chain of receipts
Every signed package links to the one before it and the one after it. You can look at the whole chain of calls: what was asked, by whom, using which data set, and in what order.

Nine checks.
Every one has to pass before the model runs.

Checks 1 to 6 come from the base SHOD stack. Checks 7 to 9 are HiveBound's own pre-commitment checks. They run cheapest first, and stop the moment one fails.

1 ALLOWLIST Checks the model and provider are on the approved list SHOD
2 RATE_LIMIT Checks the agent isn't calling faster than its allowed rate SHOD
3 TOKEN_BUDGET Checks max_tokens stays inside the approved limit SHOD
4 SCOPE_CHECK Checks the passport has the right scopes and the intent category is approved SHOD
5 EXPIRY Checks the passport and intent haven't expired SHOD
6 SIGNATURE Checks the Ed25519 signature matches the passport's public key SHOD
7 BOND_PRESENT Checks the money bond fields are filled in: amount, currency, chain HiveBound
8 BOND_SUFFICIENT Checks the bond meets the minimum and the currency is allowed HiveBound
9 INTENT_MATCH Checks the intent fields are complete and match the passport HiveBound

Five steps.
One bonded AI call.

01
Build the package
Identity, intent, and bond get put together, formatted with RFC 8785 JCS, and signed with Ed25519. The package ID comes from a SHA-256 fingerprint.
02
Nine checks
BondValidator runs all nine checks in order, cheapest first. If any one fails, the call stops right there. No model call, no charge.
03
Call the provider
The request goes out to any OpenAI-compatible endpoint. The package ID rides along in the header so the provider can match it up.
04
HAHS receipt
hivemorph closes out the receipt. The answer is locked to its data set and shows any tampering. DR5 signs the data reference. Anyone can check it on their own computer, forever.
05
x402 settlement
The bonded money settles through x402 on Base (chain ID 8453), in USDC, with a 5bps treasury fee. A settlement reference comes back in the response.

Drop it in. Wrap any provider you already use.

HiveBoundAdapter wraps any OpenAI-compatible endpoint. You keep your same provider. Your calls just become contracts.

TypeScript · HiveBoundAdapter @hive/hivebound v1.0.0
import { HiveBoundAdapter, buildPassport } from '@hive/hivebound';

// Build the agent's passport from its Ed25519 private key
const passport = buildPassport(PRIVATE_KEY, {
  agent_label: 'legal-agent-01',
  tenant_id:   'acme-corp',
  scopes:      ['inference', 'extraction'],
});

// Wrap any OpenAI-compatible provider
const adapter = new HiveBoundAdapter({
  provider_base_url: 'https://api.fireworks.ai/inference/v1',
  provider_api_key:  FIREWORKS_KEY,
  privateKeyHex:     PRIVATE_KEY,
  passport,
  bond_config: {
    amount_raw:   '1000',   // 0.001 USDC
    currency:     'USDC',
    chain:        'base',
    chain_id:     8453,
    treasury_bps: 5,
    x402_channel: null,
  },
  hivemorph_url: 'https://receipts.thehiveryiq.com',
});

// Every call is now a bonded, signed contract
const response = await adapter.chat({
  model:    'accounts/fireworks/models/llama-v3p1-70b-instruct',
  messages: [{ role: 'user', content: transcript }],
  max_tokens: 1024,
});

console.log(response.envelope_id);      // the signed package ID
console.log(response.hahs_receipt_id);  // the receipt tied to this data set
console.log(response.settlement_tx);    // the x402 payment on Base

If something goes wrong,
what can you actually prove?

Scenario Without HiveBound With HiveBound
Compliance audit: "what did the model actually say?" Just a log file. It can be edited. No signature. An Ed25519 receipt you can check on your own computer. Tampering shows up.
Someone disputes the output in a lawsuit You can't prove anything. A signed record, locked to its data set, with the chain of custody intact.
Selling into a regulated company: HIPAA, SOX, EU AI rules You can't show where the model's data came from. A full chain of custody: agent identity, approved scope, and money on the line.
An agent acts on data that's been tampered with You'd never know. DR5 catches it: a detection rate of 1.000 versus 0.000, across 500 trials.
A chain of agents: who called what? No way to trace it. Every signed package links to the one before and after it. You can trace the whole chain.

Keep the speed of your provider.
Get the proof from Hive.

HiveBound snaps into any AI stack you're already running. Your calls keep flowing the same way. They just show up as contracts now.

View Signed Benchmarks Mint an Agent Capitolare
Patent Pending. Hive / TheHiveryIQ. All rights reserved
Private by design. Hive does not store your prompts. Every request is already receipted by a one-way SHA-256 fingerprint, not the words. Proof, not surveillance.