index.md

aeon - the most autonomous agent framework

aeon is an open-source, MIT-licensed autonomous agent framework. Give it a direction and it leverages 81 skills to get the work done, running unattended on GitHub Actions with self-healing loops, reactive triggers, and persistent memory. No approval loops. No babysitting.

Not affiliated with AEON, the agentic-economy payments network, or with aeon-toolkit, the Python time-series machine-learning library.

What aeon is

aeon is a framework you fork and run, not a hosted API or SaaS. It is a Next.js dashboard plus a set of GitHub Actions workflows. There is no public REST endpoint to call - the unit of work, and of integration, is a skill: a SKILL.md file with a schedule and explicit when-to-use guidance.

Two files define an agent:

Every skill reads both before it acts. Each run is scored 1-5 by Claude Haiku, and skills that fail repeatedly are diagnosed and patched automatically.

When to use aeon

Reach for aeon when an operator wants recurring, unattended work done without babysitting an interactive agent:

Get started

Paste one line into any coding agent (Claude Code, Codex, Cursor):

read https://www.aeon.fun/skills/aeon.md and follow the instructions to set up your aeon agent

That hosted skill turns the agent into an operator that forks aeon, authenticates a model, wires a notification channel, and schedules the first skill. The same one-line command is on the homepage.

Machine-readable resources

The reachable API is v1 (URL-path versioned: https://www.aeon.fun/v1/ask; API-Version header). Responses carry standard RateLimit headers; /ask returns 429 with Retry-After over the soft limit. An aeon instance is driven by its own ./aeon CLI (in the forked repo), which also ships as a Claude Code plugin.

Documentation

Source & company