Summary

Helios is an autonomous research agent inspired by Andrej Karpathy's 'autoresearch', designed to automate complex research tasks like training machine learning models. It targets researchers and power users, offering a hands-off approach to experimentation, iteration, and result generation, even across multiple SSH-connected machines. The primary benefit is enabling users to leave complex, multi-step research processes running unattended, waking up to completed goals.

How It Works

Helios operates an autonomous loop: understanding a goal, breaking it into experiments, launching them via remote_exec_background with live metric parsing and monitoring, comparing runs, and iterating until the goal is met. It leverages a persistent memory system, storing experiment configurations, metrics, and observations as a virtual filesystem. Seamless SSH integration allows it to manage workloads on remote machines, forwarding all operations and using local machines for lighter tasks. Key components include metric tracking, a memory tree, and composable skills.

Quick Start & Requirements

• Installation: npm install -g @snoglobe/helios
• Prerequisites: Node.js 20+
• Authentication:
  • Claude: Install Claude CLI (claude login) or set ANTHROPIC_API_KEY.
  • OpenAI: OAuth login on first run (requires ChatGPT Plus or Pro).
• Remote Access: SSH configuration for remote machines.
• Links: No specific quick-start or demo links provided in the README.

Highlighted Details