aaronjmars/aeon

What it solves

Aeon addresses the need for persistent, unattended AI agents that can perform complex, recurring workflows. Unlike many agent tools that require constant human approval for every step, Aeon is designed to run autonomously on a schedule, handling tasks like feature development, security vulnerability scanning, and market research without constant babysitting.

How it works

Aeon operates using "skills," which are Markdown files containing a prompt and configuration metadata. These skills are executed via a scheduler (using cron syntax) and can be triggered manually via a CLI or dashboard. The framework uses a self-healing loop where an LLM (like Claude) scores every run; if a skill fails or produces low-quality output, specialized skills like skill-repair and self-improve automatically diagnose and patch the issue. It can also spawn specialized instances of itself to manage larger fleets of agents.

Who it’s for

• Developers who want to automate repetitive tasks like PR reviews, changelog generation, or prototype deployment.
• Security Researchers looking for automated, responsible vulnerability scanning and disclosure.
• Crypto/Market Analysts needing continuous monitoring of tokens, DeFi protocols, or market narratives.
• Productivity Seekers wanting automated digests and notifications via Telegram, Discord, or Slack.

Highlights

• Autonomous Self-Healing: Automatically detects, diagnoses, and repairs failing skills using LLM-based scoring.
• Extensible Skill System: New capabilities are added simply by creating Markdown files—no complex plugin API required.
• Multi-Model Support: Connects to Claude (via subscription or API), OpenRouter, and various other LLM gateways.
• Production-Ready Capabilities: Capable of shipping code to repos, deploying web apps to Vercel, and performing deep research.
• Fleet Management: Can spawn and manage multiple specialized instances of itself to distribute workloads.