Aeon vs Claude Code: same engine, nobody at the wheel

Aeon vs the Others, #2. Claude Code is a session you drive; Aeon puts that session on cron.

This is a stack, not a rivalry. Claude Code is Anthropic's interactive coding agent: you sit in the terminal, steer it turn by turn, review what it does. Aeon is an open-source framework that runs agents unattended on GitHub Actions, and Claude Code is one of the ten engines it can drive. So the real question isn't which one is better. It's whether anyone needs to be watching.

If you're at the keyboard, Claude Code alone is plenty. If the work should happen while you're not, that's what Aeon is for.

Set up Aeon. You can ask Claude Code itself to do the setup.

Claude Code is an interactive agent in your terminal (and IDE, and desktop). It reads your repo, edits files, runs tests, opens PRs, with you there to direct and approve. Session by session, you drive the ops.

Aeon takes the same kind of agent and removes the operator. Skills are plain SKILL.md files that run on GitHub Actions, on a schedule or on reactive triggers, through any of ten harnesses (Claude, Grok, Codex, Pi, Vibe, Kimi, fx, Cursor, Hermes, GLM). A model scores every run 1-5. A repair loop (heartbeat, skill-health, skill-repair) patches skills that break. Output goes to Telegram, Discord, Slack, Buzz, or email.

• Runtime - Aeon: GitHub Actions, no machine needed. Claude Code: your terminal / your machine.
• Operator posture - Aeon: configure once, walk away. Claude Code: you drive each session.
• Stays on without you - Aeon: yes, cloud, scheduled. Claude Code: no, needs your terminal open.
• Scheduling - Aeon: native cron per skill. Claude Code: you invoke it.
• Reactive triggers - Aeon: fires on conditions. Claude Code: none.
• Self-healing - Aeon: auto-diagnoses and patches failing skills. Claude Code: none.
• Quality scoring - Aeon: every run scored 1-5 by a model. Claude Code: none.
• Skill format - Aeon: Markdown (SKILL.md). Claude Code: Markdown skills plus MCP.
• Persistent memory - Aeon: versioned files (SOUL.md, STRATEGY.md). Claude Code: session plus CLAUDE.md.
• Model choice - Aeon: 10 harnesses, swap in config. Claude Code: Anthropic models.
• Interactive pairing on code - Aeon: not the point. Claude Code: what it's built for.
• Fine-grained human control - Aeon: optional gate (Fleet Watcher). Claude Code: approval loop by design.
• Fleet of specialized instances - Aeon: spawn-instance plus fleet-control. Claude Code: none.
• Hosting cost - Aeon: free on public repos. Claude Code: your machine.

Aeon: free (MIT). Actions minutes cost nothing on public repos. You bring the model auth, including the same Claude subscription you already use for Claude Code.

Claude Code: included with a Claude subscription (Pro $20/mo, Max from $100/mo) or pay-as-you-go API.

If you already pay for Claude, Aeon adds roughly nothing. It reuses that auth and runs on free compute.

• You're actively building and want an agent pairing with you.
• The task is one-off, exploratory, or too fuzzy to schedule.
• You want to approve every diff before it lands.

• You keep running the same session every day (morning digest, dependency check, repo triage, PR review) and it should just be a cron entry.
• You want work to continue when the laptop is closed.
• You want failures caught by a repair loop instead of by you noticing something didn't run.
• You want the option to route some skills to other models (Grok, Codex, Kimi...) without rewriting anything.

If you need an agent you watch, use Claude Code. If you need an agent that watches itself, use Aeon.

Does Aeon replace Claude Code?

No, it drives it. Claude Code is one of Aeon's ten harnesses. Most setups use both: interactive sessions at the desk, Aeon skills on cron for the rest.

Can't I just script Claude Code with cron myself?

You can, and that's roughly where Aeon started. But then you're building the scheduler config, notification channels, per-run quality scoring, the health and repair loop, memory files, guardrails, a dashboard, and fleet management yourself. Aeon is that layer, already built, MIT-licensed.

Do my existing Claude Code skills work in Aeon?

Same idea in both: a Markdown file with frontmatter and a prompt. Porting is usually copy, adjust frontmatter, schedule. There's also a create-skill skill that writes new ones from a sentence.

What about safety without a human approval loop?

Read-only skills can't write to the repo, irreversible actions fail closed, and secrets never touch the command line. If you still want a human gate, the optional Fleet Watcher layer approves every run before it executes.

How do I set it up from Claude Code?

Tell it: read https://www.aeon.fun/skills/aeon.md and follow the instructions. It forks the repo, wires auth and a notification channel, and schedules your first skills.

• #1 - Aeon vs Hermes: two autonomous agents, one big difference
• #3 - Aeon vs ChatGPT: the agent that works while you're gone
• #4 - Aeon vs Grok Bot: the $300 agent and the free one

Fork aeonfun/aeon, run ./aeon, pick skills, walk away.

No approval loops, no babysitting!