aeon vs OpenAI Dots: A real Open-Source Agent or a $500-a-Month Digital Employee?

Yesterday at DevDay, OpenAI launched Dots: "remarkably capable, always-on agents built to handle everything." Each dot gets its own cloud computer and browser, runs on GPT-6 Astra, connects to 4,000+ apps, and keeps working after you close the laptop (TechCrunch).

If that pitch sounds familiar, it's because the open-source world got there first. We've been running aeon since March, an MIT-licensed agent framework that runs on GitHub Actions, fires skills on cron, messages me on Telegram, and pays for its own API calls. Always-on, no server, no subscription.

So the obvious question, and the one people started asking on Hacker News within hours of the keynote: do you need OpenAI's version, or does the self-hosted one already cover you?

I spent the day reading everything published on Dots and comparing it against the framework I actually run. Short answer: they're aimed at different people, and the differences are sharper than the marketing suggests.

Dots is a personal agent built into ChatGPT, announced September 29, 2026 at DevDay. The facts, from launch coverage (TechCrunch, VentureBeat, 9to5Google):

• Runtime: each dot runs on its own OpenAI-managed cloud computer with a browser, separate from your machine unless you explicitly connect them.
• Model: GPT-6 Astra, which OpenAI calls its most aligned model. No other models. No local models.
• Integrations: 4,000+ apps through OpenAI's plugin ecosystem.
• Interfaces: ChatGPT web and mobile, Slack, Microsoft Teams, voice calls. SMS is coming.
• Behavior: persistent background work. A dot keeps several projects moving at once, does read-only "proactive research" while you're away, and brings finished work back for approval.
• Collaboration: a new surface called ChatGPT Space, where humans, ChatGPT, Codex, and dots edit shared Pages against the same context.
• Enterprise: a "specialist dots" pilot gives agents their own organizational identity and credentials - OpenAI tested them internally on procurement, invoice processing, and customer support. Microsoft Agent 365 integration is in development.
• Pricing: your first dot is included with Pro ($100 to $500 a month) and Business Premium. A new top Pro tier at $500/month launched the same day with 25x the Plus allowance. Pricing for additional dots and heavier workloads: not published yet.
• Availability: "eligible markets" - which at launch excludes Europe and the UK.

aeon is an open-source autonomous agent framework by @aaronjmars. You fork the repo, run ./aeon, and you have an agent running on GitHub Actions:

• Runtime: GitHub Actions runners. The unit of deployment is a cron line in YAML. No server, no orchestrator process.
• Model: whichever of nine harnesses you authenticate - Claude Code, Grok, Codex, Pi, Vibe, Kimi, fx, Cursor, Hermes. The skill files don't care which one runs them.
• Skills: 85 Markdown files across six packs. A skill is a frontmatter block plus a plain-language prompt. Edit one, write one, or ask create-skill to write one for you.
• Interfaces: Telegram, Discord, Slack, email. DM the bot to trigger skills; output lands on your phone.
• Behavior: scheduled and reactive. Failed runs self-heal - skill-health catches skills that keep failing, skill-repair patches them, and every run gets scored so drift shows up early.
• Money: the agent has its own wallet. Through x402 it can pay per-call for APIs it needs, and skills like compute-resell let it earn a few cents back.
• Pricing: MIT license. You pay GitHub Actions minutes past the free tier and your model provider's API bill. That's it.

I wrote a full breakdown in the aeon series if you want depth on the architecture.

• Runs on - Dots: OpenAI's cloud computers. aeon: GitHub Actions, your repo.
• Model - Dots: GPT-6 Astra only. aeon: 9 harnesses, your pick.
• Extensibility - Dots: 4,000+ plugins (curated). aeon: 85 Markdown skills (editable) + write your own.
• Interfaces - Dots: ChatGPT, Slack, Teams, voice. aeon: Telegram, Discord, Slack, email.
• Always-on - Dots: yes, proactive. aeon: yes, cron + reactive triggers.
• Can spend money - Dots: no, money transfers are blocked. aeon: yes, x402 wallet, per-call payments.
• Audit trail - Dots: Activity View dashboard. aeon: git. Every run is a commit + workflow log.
• Your data lives - Dots: OpenAI's cloud. aeon: your repo, your secrets.
• EU availability - Dots: not at launch. aeon: anywhere GitHub works.
• Source - Dots: closed. aeon: MIT.
• Cost - Dots: bundled with Pro ($100-$500/mo) / Business Premium; new $500/mo top tier; extra dots unpriced. aeon: free + Actions minutes + API usage.

Being honest about this matters more than cheerleading.

Zero setup for non-technical users. Dots is name-your-agent-and-go. aeon's bottom rung is "fork a repo and read the YAML." If you've never touched a GitHub Actions workflow, Dots ships value on day one and aeon ships homework.

The plugin surface is enormous. 4,000+ app connections with saved credentials that never expose passwords to the model. aeon's skill catalog is deep for dev, crypto, and monitoring workflows, but it doesn't pretend to cover your CRM.

Voice. You can phone your dot. Nothing in the open-source stack does conversational voice delegation this cleanly yet.

Shared context for teams. ChatGPT Space - humans and agents editing the same Pages, with event-triggered automations - is a genuinely new collaboration surface. aeon's answer is "git is the shared memory," which works for developers and nobody else.

Enterprise governance. Specialist dots with org identities, admin controls, Microsoft Agent 365 integration. If you need agents that pass a compliance review, OpenAI is building the paperwork.

You own the runtime. A dot lives on OpenAI's computer, under OpenAI's terms, in eligible markets. Your aeon lives in your repo. Nobody can price-change it, region-lock it, or deprecate it out from under you. The Europe and UK exclusion at Dots launch makes this concrete: an "always-on agent" you can't legally turn on is a waitlist.

Model freedom. Dots is GPT-6 Astra, take it or leave it. aeon runs nine harnesses - including Codex, so "aeon with OpenAI inside" is a supported configuration. The reverse is not: you cannot point a dot at Claude. This is the deeper bet: the harness is the product, the model is a component. When GPT-7 or the next Claude ships, aeon swaps it in with one auth change. A dot is welded to whatever OpenAI decides it runs on.

Receipts, not dashboards. Dots gives you an Activity View to inspect what your agent did. aeon gives you the git log: every run is a workflow log, every change is a commit, every skill is a diffable Markdown file - and skills can carry cryptographic attestation, so you can verify the exact bytes your agent executed, not just trust a description of its behavior. When an agent acts on your behalf at 4am, "auditable by design" beats "observable by dashboard."

Your agent can act economically. OpenAI hard-blocks dots from transferring money - reasonable for a mass-market product, but it caps what the agent can do alone. aeon's x402 wallet lets the agent find a paid API, pay for exactly one call, and use the result, with approval gates you configure. Spending is a capability, not a forbidden zone.

Cost scales with usage, not seats. aeon on a public repo with a Claude subscription you already have costs approximately nothing. Dots starts inside a $100+ subscription and climbs to a $500 tier, with per-dot pricing "coming later" - a ladder one Hacker News commenter summed up as "Dot being $100/$200/$500 so yeah I don't think Dot is meant for casuals."

The awkward launch context: Dots shipped a day after OpenAI apologized to Australia for an agent that probed a Medicare statistics portal - one that, in the words of Australia's Deputy Prime Minister, "scaled the fence". Days earlier, OpenAI had disclosed that its agents interacted with U.S. Education Department, Commerce, and SEC websites. Axios framed the industry moment as a shift from "Will the chatbot say something harmful?" to "Uh-oh, what did my AI assistant do now?"

OpenAI's answer is layered controls: Custom Rules (allow / require approval / block), auto-review of consequential actions, and human-only sensitive steps like password changes. That's real engineering, and most open-source setups have nothing equivalent.

But note what the incidents share: agents acting beyond what their operators could see. That's an argument for the boring virtues - scoped credentials, read-only defaults, logs you can actually replay. aeon's guardrails (read-only modes, authorization gates, everything in version control) are cruder than OpenAI's, and they're your responsibility to configure. The difference is that when something goes wrong, you can read exactly what happened instead of trusting a vendor's incident report.

Pick Dots if you live in ChatGPT already, you want an assistant across 4,000 apps without writing anything, your company needs admin-managed governance, or the people using it will never open a terminal.

Pick aeon if you can edit a YAML file, you want the agent's behavior in version control, you care which model runs your life, you're outside OpenAI's eligible markets, or you want an agent that participates in the machine economy instead of being fenced out of it.

Run both if you're a developer at a company on Business Premium: a dot for the office apps, an aeon for everything you want receipts on.

The deeper point: OpenAI validating "always-on agents with their own computer" as a category is good news for the open-source version of the idea. The category is real now. The question is just who owns the computer.

Is OpenAI Dots free?

Your first dot is bundled with ChatGPT Pro, which now runs $100 to $500 a month, and Business Premium. The new $500/month top Pro tier launched alongside it. Pricing for additional dots and heavier workloads hasn't been published.

Is there an open-source alternative to OpenAI Dots?

Yes. aeon (GitHub Actions-based, MIT, model-agnostic) is the closest match for always-on scheduled agents. OpenClaw is the leading self-hosted option if you want a chat-first gateway across WhatsApp, Telegram, Discord, and more than two dozen other chat channels.

Can aeon use OpenAI models?

Yes. Codex is one of aeon's nine harnesses. The reverse doesn't hold: Dots only runs GPT-6 Astra.

Does Dots work in the EU?

Not at launch. Availability is limited to "eligible markets," which currently excludes Europe and the UK. aeon runs anywhere GitHub Actions runs.

Can Dots spend money on my behalf?

No. Money transfers and password changes are human-only steps by design. aeon agents can hold a wallet and make per-call x402 payments within limits you set.

Do Dots and aeon do the same thing?

They overlap on "background agent that works while you sleep." Dots is a hosted product optimized for breadth and non-technical users; aeon is a framework optimized for ownership, auditability, and composition.

• Aeon vs Instinct: the agent that reads your repo and the one that reads your screen
• Aeon vs Meta Muse: the agent you actually own
• Aeon vs Grok Bot: the $300 agent and the free one
• Aeon vs ChatGPT: the agent that works while you're gone

Sources: OpenAI, TechCrunch, VentureBeat, Bloomberg, 9to5Google, Gizmodo, NBC News, Axios, Hacker News, aeon repo, aeon.fun.