Ashlar Conductor™
The high-throughput, hardware-attested micro-clearinghouse for autonomous AI inference streams. Meter tokens per second, slice sub-agent budgets non-interactively, and automatically reclaim unspent capital on stream termination.
The Problem in Autonomous AI Inference: In conventional LLM and API infrastructure, payments are either billed post-hoc on credit risk or charged upfront via coarse credit bundles. For autonomous AI agent swarms executing thousands of continuous multi-turn completions per minute, upfront prepayments lock up idle liquidity while post-hoc billing exposes model providers to insolvency and unpaid computing debts. Ashlar Conductor™ solves this impasse by introducing cryptographic streaming micro-clearing—metering compute at the individual token level and settling costs progressively as tokens transit the wire.
How Two-Phase Micro-Clearing Operates: When an enterprise client or AI agent begins an inference task, it opens an x402.compute-session/v0 channel by locking a finite budget ceiling (e.g., $0.50 RLUSD) through a gasless off-chain signature (EIP-3009 transferWithAuthorization or XRPL conditional escrow). As the confidential inference server (hosted inside an Intel TDX hardware enclave) generates output tokens, it issues sequential cryptographic vouchers back to the client. Each voucher monotonically updates the cumulative spend without requiring on-chain network consensus or paying per-token blockchain transaction fees.
Guaranteed Capital Reclamation & SCITT Accountability: If an agent aborts an inference stream early—such as when a search tool returns an early answer or an error occurs—the session closes immediately. The exact token consumption is finalized against the hardware-signed voucher, and 100.0% of unspent capital is automatically returned to the payer's wallet. Conductor records the final signed accounting transcript directly into an immutable SCITT (Supply Chain Integrity, Transparency, and Trust) audit envelope, providing cryptographic proof of exact delivery for compliance auditors and enterprise CFOs.
- Click "Start AI Stream" below to launch an active inference streaming session.
- Watch the real-time meters track token emission, settled cost, and remaining budget.
- Click "Close & Reclaim Refund" at any point to halt inference and trigger instant unspent capital refunding.
- Zero Idle Capital: Never deposit lump sums; authorize only the max ceiling per prompt.
- Sub-Millisecond Verification: Intel TDX vouchers verify in 0.08 ms (p50), outperforming HTTP roundtrips.
- Dual Settlement Rails: Supports Flare Network (FCUSD/FCYD) and XRP Ledger (RLUSD) interchangeably.
Hardware Enclave & SCITT Verification
Institutional Scaling Curve
- 250 Concurrent Autonomous Workers: Deployed across 5 specialized squads (Oracle Ingest, Arbitrage, Risk Gate, Settlement, SCITT Auditor).
- Sub-Millisecond Verification: 2,500 monotonic vouchers verified in 341.6 ms (7,318 vouchers/sec at 1,802,578 tokens/sec with 0.08 ms p50 latency).
- Single-Core Scale: Swarm throughput scales to 1,932,694 tokens/sec at 500 agents (7,751 vouchers/sec in 645.1 ms).
- Multi-Core Worker Pool: Reaches 6,948,103 tokens/sec (27,792 vouchers/sec in 179.9 ms) across parallel 8-core worker threads.
- Hardware Enclave Security: Enforced under live Intel TDX hardware attestation with 100.0% unspent budget reclamation.
SDK Integration (Zero Dependencies)
import { ConductorGateway } from '@ashlar-blue/x402-trust';
// 1. Initialize streaming session with budget cap ($0.50 RLUSD)
const gateway = new ConductorGateway();
const { challenge, state } = gateway.initSession({
sessionId: 'sess-enterprise-ai-01',
payer: '0xPayerWallet...',
payee: '0xFacilitatorNode...',
tokenRatePicoUSD: 2000000n, // $0.000002 / token
secondRatePicoUSD: 50000000n, // $0.00005 / sec
maxSpendPicoUSD: 500000000000n, // $0.50 Max Cap
asset: 'RLUSD',
publicKeyPem: clientPublicKey
});
// 2. Stream tokens & submit incremental vouchers
const result = gateway.processChunk(sessionId, tokens, elapsedMs, signedVoucher);
// 3. Finalize & reclaim unspent balance (Deterministic Auto-Reclamation)
const verdict = gateway.closeSession(sessionId);
console.log(`Settled: $${verdict.amountSettledPicoUSD}, Refunded: $${verdict.refundAmountPicoUSD}`);