Gladiator
runtimenoteslabs/gladiator
Hackathon project that pits two Hermes Agent and Paperclip AI companies against each other
Gladiator is a Python demo from the Nous Research Hermes Agent Hackathon (March 2026) in which two autonomous AI companies, Blitz and Craft, compete on identical starter repos, using Paperclip for orchestration and Hermes Agent as the workers.
What Gladiator does
Gladiator runs two zero-human companies on the same product, an llm-judge starter repo. Blitz follows a growth-first strategy and Craft a quality-first one. Paperclip handles orchestration, and Hermes Agent workers do the tasks. In the README's demo flow, 9 agents complete 10 tasks in about 5 to 6 minutes while creating skills, growing memory and writing code, and the winner is declared by projected GitHub stars.
A live dashboard built with FastAPI, server-sent events and vanilla JavaScript visualizes the battle in nine sections, including the scoreboard, task boards, code comparison, Gantt chart, audit trail, learning evidence and merge controls. Evidence is stored in SQLite and a Python watcher detects completion. After the match the two companies merge and skills transfer between teams, which the project presents as evidence that Hermes learning is real. Paperclip connects to Hermes through the hermes-paperclip-adapter npm package.
Key features
- Two rival companies, Blitz and Craft, with different strategies
- Paperclip orchestration with Hermes Agent workers via hermes-paperclip-adapter
- Live dashboard on port 4000 built with FastAPI and server-sent events
- SQLite evidence database and a Python watcher that detects completion
- Merge step that transfers skills between the two companies
When to use it
- Studying how Hermes Agent skills and memory grow during an autonomous multi-agent run
- Using the setup guide as a worked example of wiring Hermes Agent into Paperclip
- Demonstrating competing agent strategies on a live dashboard
Who it is for: Developers who want a worked example of Hermes Agent workers running inside Paperclip companies.
How it fits with Hermes Agent
Uses Hermes Agent as the worker runtime inside Paperclip, and its setup guide starts with the official Hermes installer.
Requirements: Python 3.11+, Node.js 20+, pnpm 9+, PostgreSQL 16+, an Anthropic API key, and roughly $5-6 USD per 10-minute run
Note: It is a hackathon demo last updated in March 2026, and each run needs an Anthropic API key and costs roughly $5 to $6.
FAQ
What is Gladiator?
Gladiator is a hackathon project in which two autonomous AI companies, Blitz and Craft, compete to maximize GitHub stars on identical starter repos. It uses Paperclip for orchestration and Hermes Agent workers.
Does Gladiator work with Hermes Agent?
Yes, Hermes Agent supplies the workers. Paperclip reaches them through the hermes-paperclip-adapter package, and the README's setup guide installs Hermes first and configures the Anthropic provider.
What do I need to run Gladiator?
You need Python 3.11+, Node.js 20+, pnpm 9+, PostgreSQL 16+ and an Anthropic API key. The README estimates about $5-6 USD per 10-minute competition run with 9 Sonnet agents.
Similar multi-agent for Hermes Agent
All multi-agentRun a bounded, supervised team of Hermes Agent workers on one host, coordinated through Kanban
gutomec Nirvana-OSRuntime-agnostic multi-agent engine that runs companies, squads and mind-clones from one plain-prose brief
Jovancoding Network-AITypeScript multi-agent orchestrator with a locked shared blackboard, guardrails and 32 framework adapters
forcewake Hermes ConductorPatterns for running Claude Code, Codex CLI and other coding agents under one Hermes kanban controller
heggria taskflowDeclarative DAG runtime for coding-agent workflows with verification, resume, replay and effect checks
Abruptive Ankh.mdPer-folder scoped Hermes Agents with their own config, skills and memory, launched by the hermes command
Related guides: Build agent teams with the Hermes kanban board · Run multiple Hermes agents with profiles · Connect Hermes agents on several machines with Hermes Desktop