We engineer the
verifiable compute stack
for the EVM.

Humbleaf builds the cryptographic infrastructure required for smart contracts to natively verify autonomous AI execution. No centralized trust. Just pure math.

5
Audited Contracts
<500ms
Verification Latency
1000×
Cheaper than zkML
20%
Protocol Fee

One Stack. Complete Verification.

Three interlocking protocol layers that form a single, cohesive physics engine for trustless AI execution on the EVM.

LAYER 1

PRIOR

The Data Availability Anchor

Bypasses the Ethereum calldata bottleneck. PRIOR securely routes and anchors gigabyte-scale enterprise context windows to IPFS, settling only a 32-byte cryptographic proof on Base L2.

∞ GB
Payload
32 bytes
On-chain cost
Base L2
Settlement
LAYER 2

AgentChain

The opML Execution Engine

Reads the PRIOR payload, executes the AI inference deterministically inside our proprietary fault-proof environment, and settles the execution trace on-chain — secured by dynamic slashed bonds.

Deterministic
Execution
$5k Bonds
Security
5 Audited
Contracts
LAYER 3

The Supply Cartel

DePIN Compute Network

Powered by a decentralized network of sunk-cost European GPU miners utilizing compute multiplexing for zero-opportunity-cost inference.

DePIN
Supply
EU GPU
Nodes
≈ $0
Marginal cost

10 Lines of Code to
Cryptographic Trust.

We abstracted the brutal complexity of deterministic execution and IPFS data routing into a single NPM package.

Pass a 2GB context payload. Get back a mathematically verified opML State Root. That's it.

The Stripe of verifiable AI.
verify.ts
import { AgentChain } from "@agentchain/sdk"

const client = AgentChain.init({
  network: "base-mainnet",
  apiKey: process.env.AC_KEY
})

const result = await client.execute({
  model: "llama-3-70b",
  contextPayload: corporaBuffer, // 2 GB via PRIOR DA
  prompt: "Analyze and summarize."
})

// result.stateRoot → 0x7f3a...b91c
// result.verified  → true
// result.txHash    → 0x91ca...d3f7

1000× Cost Reduction.

zkML is a tarpit. We chose the only architecture that actually scales.

MetriczkMLAgentChain opML
Cost per inference$5,000+$0.05
Verification typeZero-Knowledge ProofOptimistic ML (opML)
Proof generationHours to days< 500ms
Model supportTiny models only70B+ parameter models
GPU requirementSpecialized ZK hardwareConsumer GPUs (DePIN)
Security modelCryptographic proof$5k slashed bonds + dispute
20%
Mathematically enforced protocol fee on all compute routed through the EVM. Collected at the smart contract level. No opt-out.
$0.05
Per inference — including PRIOR data routing, deterministic execution, and on-chain settlement. All-in.
Live on Base Sepolia

Proof of
Physics.

Real-time pipeline: autonomous agents generate PRIOR payloads, AgentChain ingests and executes deterministically, workers process and return cryptographic state roots — verified on-chain in under 500ms.

PRIOR hash anchored to IPFS
Deterministic opML execution
On-chain state root settlement
Humbleaf

Humbleaf is an infrastructure R&D company. In a single 5-week sprint, we engineered a deterministic opML execution engine, built a custom IPFS data routing protocol, and deployed 5 audited smart contracts to Base L2 to solve the hardest problem in crypto: trustless AI execution.

We don't write whitepapers. We ship physics.