╔══════════════════════ SYSTEM // ONLINE ══════════════════════╗
║ ID : KoNit ║
║ MODE : BUILD · BREAK · SECURE · DECENTRALIZE ║
║ FOCUS : AI AGENTS · SECURITY · WEB3 · AI × FINANCE ║
║ SIGNAL : MAJOR CONTRIBUTOR · HERMES AGENT ║
╚════════════════════════════════════════════════════════════════╝
I build and research systems at the intersection of AI agents, agent security, decentralized infrastructure, and financial protocols. My background is in Web3 and blockchain systems; my current work is increasingly focused on reliable agent infrastructure, adversarial evaluation, multi-agent systems, and AI × finance / Web3.
Major contributor to NousResearch/hermes-agent, contributing across agent infrastructure, Desktop, reliability, tooling, and cross-platform behavior.
OFFICIAL RELEASE CREDITS
- Hermes Agent v0.22.2 — contributed to Doctor diagnostics and dependency security, improving npm vulnerability remediation guidance.
- Hermes Agent v0.22.1 — contributed to Doctor security checks and dependency management, clarifying how npm audit findings should be addressed.
- Hermes Agent v0.22.0 — The Threshold Release — co-credited for Stable release channels (stable updates and rollback safeguards), and credited in relay and Claude rate-limit improvements. Listed among the Top Community Contributors (8 merged PRs in this release cycle).
- Hermes Agent v0.21.2 (v2026.9.11) — 13 authored commits in the v0.21.1 → v0.21.2 release window, including 3 directly merged PRs; credited among the release's 140 contributors:
- Desktop and Bot Mode: #107239 prevents fixed titlebar controls from overlapping extension-provided pages; #106860 scopes group-chat
(you)labels to the correct connection; remote liveness accounts for cold-start probes.- Installer and runtime reliability: #107186 stamps the macOS setup app with the release SemVer instead of a stale placeholder; shell PATH probes terminate their child on timeout.
- State and tooling: FTS recovery diagnostics identify blocked rebuilds; TUI model config rejects sessionless mutation; Kanban tools preserve explicit scratch/project selection.
My earlier work and research centered on blockchain and decentralized systems. I continue to work across Web3 infrastructure, DeFi, smart contracts, and financial mechanisms, while moving deeper into AI-agent systems and agent security.
I am particularly interested in problems where these areas overlap: autonomous agents interacting with financial systems, secure tool-using agents, decentralized execution, and mechanism-aware AI systems.
Systems: LLM Agents · Multi-Agent Systems · Agent Security · EVM / DeFi · Smart Contracts · Distributed Systems · Wallet Infrastructure
| Area | Current direction | |
|---|---|---|
| 🤖 | AI Agents & Agent Infrastructure | Orchestration, multi-agent coordination, tool use, runtime reliability, and local/cloud agent systems |
| 🛡️ | Agent Security | Red teaming, adversarial evaluation, robustness, and safe credential / execution boundaries |
| ⛓️ | Web3 Systems | Wallets, smart contracts, DeFi/AMM systems, cross-chain infrastructure, and onchain financial applications |
| ◈ | Research | Mechanism design, game theory, and AI × decentralized / financial systems |
Automatic multi-agent red-teaming framework for evaluating AI-agent robustness.
Attacker, target, and evaluator agents operate in an iterative adversarial feedback loop, with configurable LLM backends and distributed execution through the Naptha SDK.
ATTACKER ──► TARGET ──► EVALUATOR
▲ │
└────── feedback loop ◄───┘
Open to open-source engineering, AI / agent-security research, and technically serious work at the intersection of AI, Web3, and financial systems.
BUILD // TEST // BREAK // SECURE // SHIP