Aeon, the open-source always-on agent (not the OpenAI one)

If you searched "Aeon agent" this week, you're probably looking for one of two things.

The first is OpenAI's. According to reports, OpenAI is preparing a persistent, always-on agent under the name Aeon, possibly around DevDay on September 29. As of today there's no official page or documentation, just a model ID spotted in a feature flag and some reporting. We don't know more than that, and we're not affiliated with OpenAI.

The second is this one. Aeon is an open-source autonomous agent framework. The first commit landed on March 4, 2026, and it has been public since that month: 1,300+ commits, MIT licensed, and a fleet of live instances doing real work every day.

Since both will be called "always-on agents," here's what makes this one different.

Aeon is an agent that lives in your GitHub repo.

• It runs on a schedule in GitHub Actions. No server to host and no platform to sign up for. Fork the repo, turn skills on, and it runs itself.
• Its memory is committed to git. Every run reads and writes plain files in memory/. You can diff what your agent learned yesterday, revert it, or grep it.
• Skills are Markdown. A skill is a SKILL.md file, and aeon.yml says which ones run and when. Aeon ships with 85 bundled skills plus installable community packs.
• It works with any model. One harness adapter dispatches every run across Claude Code, Codex, Cursor, Grok, Kimi, and more, so you can pin models per skill and switch providers without rewriting anything.
• Secrets are least-privilege. Each skill declares what it requires:, and only those secrets get wired in.

An always-on agent accumulates context about your work: your projects, your priorities, your voice. The question is where that context lives.

With Aeon, the answer is simple: in a repo you own. The agent's strategy is a file (STRATEGY.md). Its voice is a folder (soul/). Its memory is a commit history. If you stop using Aeon tomorrow, you keep everything, in formats any tool can read.

That's the design decision we made on day one, and it's the reason Aeon runs on infrastructure you already control instead of ours.

Fork it and pin your instance:

gh repo fork aeonfun/aeon --clone && cd aeon
gh repo set-default <you>/aeon

Or install the operator console as a plugin:

• Claude Code: /plugin marketplace add aeonfun/aeon then /plugin install aeon@aeon
• Codex: codex plugin marketplace add aeonfun/aeon then codex plugin add aeon@aeon

Full setup guide: www.aeon.fun. Source: github.com/aeonfun/aeon.

Is Aeon affiliated with OpenAI?

No. Aeon is built by Aeon Inc and has been open source since March 2026.

What does it cost?

The framework is free and MIT licensed. Public forks run on GitHub's free Actions minutes; you bring your own model API keys or coding-agent subscription.

Which models does it support?

Any model reachable through a supported coding-agent CLI. Most instances run a cheap default model with stronger models pinned to the skills that need them.