Skip to content
Hyperagent: AI-powered smart contract
2026-05-14T00:00:00.000Z

Hyperagent: AI-powered smart contract

AI-powered smart contract development platform From natural language to production-ready, audited contracts in minutes.

Categoryautomation
Tags
agentaiautomateworkflow
Author
J
JustineDevs

Integrated Networks

evm
sui
solana

Responsibilities

Tags

agentaiautomateworkflow

Technologies

typescriptpythonpostgresql

My Responsibilities

  • Building From scratch
  • System designing & Architecture
  • Hardening and Polished

Description

HyperAgent is an AI-assisted smart contract development platform that turns natural language specifications into draft artifacts with automated checks, an audit workflow, simulation, and deploy preparation for SKALE Base Mainnet and SKALE Base Sepolia in the current release.

Overview

HyperAgent is an AI-powered multi-agent platform that turns natural-language specifications into draft contracts and artifacts with automated checks, audit workflow stages, Tenderly simulation, and deploy preparation on SKALE Base Mainnet and SKALE Base Sepolia for v0.1.0. The system uses a microservice and service-oriented architecture (SOA). Agents communicate via the Agent2Agent (A2A) protocol with ERC-8004 on-chain registries for identity, reputation, and validation. LLM access is BYOK: no server-side LLM keys for user workloads; keys are stored in an isolated, encrypted environment and used only for that user's runs.

Goal: Make end-to-end smart contract development verifiable, repeatable, and safe. Developers go from "idea in English" to deployed and monitored contract with as few manual steps as possible, while preserving auditability and human oversight where needed.

Features

  • Natural language to contracts – Describe behavior in plain language; get Solidity, tests, and artifacts for the audit workflow (static analysis and fuzzing).
  • BYOK – Bring your own LLM keys (OpenAI, Anthropic, Google, OpenRouter); no server-side LLM config for user workloads.
  • Tenderly simulation – Transaction simulation and reports in the pipeline before deploy when your environment and workflow enable it.
  • Security tooling – Slither, Mythril, and related tools run when the audit service is deployed and reachable; gates use findings when the audit stage completes (not a substitute for human review on high-value contracts).
  • Supported network scope – v0.1.0 is scoped to SKALE Base Mainnet and SKALE Base Sepolia. Other chains in the repo are roadmap, not current support.
  • Account abstraction and payments – ERC-4337 and EIP-7702 via Thirdweb; x402 is the intended payment wall for supported user flows, with current enforcement gaps tracked in the honesty docs.
  • Metrics and tracing – OpenTelemetry, MLflow, Tenderly monitoring, Dune dashboards where the deployment wires them.
  • RAG and memory – Curated corpora and Acontext for long-term agent memory; artifact storage via IPFS/Pinata (and compatible providers)

System Architecture Overview

HyperAgent is a layered platform. Each layer has distinct security responsibilities and trust boundaries.

┌─────────────────────────────────────────────────────────────────────────────┐
│  CLIENT LAYER                                                                │
│  Studio (Next.js) │ Wallet connect (SIWE) │ Thirdweb for deploy signing      │
└─────────────────────────────────────────────────────────────────────────────┘
                                      │
                                      ▼
┌─────────────────────────────────────────────────────────────────────────────┐
│  GATEWAY LAYER                                                               │
│  API Gateway │ JWT auth │ Upstash REST rate limits │ metering │ x402 (priced) │
│  Identity HMAC to orchestrator │ CORS │ Request validation                  │
└─────────────────────────────────────────────────────────────────────────────┘
                                      │
                                      ▼
┌─────────────────────────────────────────────────────────────────────────────┐
│  ORCHESTRATOR LAYER                                                          │
│  LangGraph pipeline │ BYOK decryption │ Agent session JWTs │ Step audit      │
└─────────────────────────────────────────────────────────────────────────────┘
                                      │
          ┌───────────────────────────┼───────────────────────────┐
          ▼                           ▼                           ▼
┌──────────────────┐    ┌──────────────────┐    ┌──────────────────┐
│  BACKEND SERVICES│    │  SECURITY TOOLS   │    │  DATA LAYER      │
│  Compile, Audit, │    │  Slither, Mythril │    │  Supabase (RLS), │
│  Simulation,     │    │  Tenderly, SCRUBD │    │  Redis, IPFS     │
│  Deploy, ROMA,   │    │  Pashov auditor   │    │  (Pinata)        │
│  Storage, Tools  │    │                   │    │                  │
└──────────────────┘    └──────────────────┘    └──────────────────┘

Layer 1: Client (Studio)

ComponentSecurity responsibility
Wallet connectSIWE (Sign-In with Ethereum) via Thirdweb; no password storage.
LLM keys (BYOK)Keys sent over HTTPS to gateway; never stored in browser after submit.
Deploy signingUser signs deploy transaction in browser; private key never leaves wallet.
API callsAll backend calls use Authorization: Bearer <JWT>; JWT from SIWE.
CORSCORS_ORIGINS restricts allowed origins; set for production domains.
Feature flagsBilling and x402 hints follow gateway env (X402_ENABLED, merchant address, metering). Studio should not assume paid routes are off in production without checking deployment config.

Studio is a thin client. All pipeline logic, BYOK handling, and safety gates live in backend services.

ToolTrust levelDeploy rulePurpose
SlitherHighBlock on High/CriticalMandatory static gate for reentrancy, access control, and related classes.
MythrilHighBlock on High/CriticalMandatory symbolic gate; findings stored with raw JSON evidence, SWC identifiers, and transaction sequences.
EchidnaMedium-highBlock when harness failsMandatory only when harness exists. Echidna is not considered successful unless HyperAgent generated or attached executable properties, assertions, or invariant functions. If no harness, return NOT_APPLICABLE.
TenderlyHighAlways block on failureMandatory predeploy execution gate; any failed simulation blocks deploy.
SCRUBDInternal-highBlock on matched patternsMandatory policy/pattern gate for curated RE/UX checks; label as internal-policy.
Pashov solidity-auditorMediumBlock only when corroboratedAlways-on reviewer; block only when corroborated by Slither/Mythril/Echidna/Tenderly or escalated by human review.
ROMAInternalNever blockSpec bootstrap and requirement shaping; not a security control.
MythXOptionalNever blockOptional compatibility backend; not a hard dependency.

Activity

Heatmap

GitHub activity is loading or unavailable. Last-known snapshots will appear here after the first refresh.

Total Contributions
0
Active Days
0
Current Streak
0 days
Longest Streak
0 days
SunMonTueWedThuFriSat
Less
More