OpenAI's dots launched on 29 September 2026 as an always-on agent that runs in the background on your behalf, and the honest read is: it's a real product, not a demo, but it's not a replacement for the automation most businesses already run. No public API, no self-hosting, and a few hard geographic gaps mean this is a consumer and enterprise-chat tool for now, not an infrastructure decision.
What Did OpenAI Actually Launch With Dots?#
At OpenAI's DevDay on 29 September 2026, the company introduced dots: a persistent, cartoon-avatar agent built on GPT-6 Astra that keeps working after you close the chat window. TechCrunch's coverage describes it as "remarkably capable, always-on agents built to handle everything," connecting to Slack, Teams, and, per OpenAI's own getting-started documentation, more than 4,000 apps through its plugin ecosystem. Your first dot is free on Pro and Business Premium plans. Enterprise access exists in beta, off by default until an admin flips it on.
Does This Change Anything If You're Not Already Automating?#
Mostly, no. If the most "automated" thing in your business is an auto-reply on your Instagram DMs, dots doesn't change your week. It's a ChatGPT feature, not a new category you need a strategy for.
If you're a SaaS founder or an ops lead already running workflows through n8n, Zapier, or a custom integration, it changes less than the headlines suggest, and that's the actual story. At WebEpex, we build that kind of automation on self-hosted n8n for SaaS founders and D2C brands across the GCC, Europe, the US, Canada, and Indian SMBs, and the first thing we checked this week was whether dots could plug into any of it. Dots has no public API as of writing, so you can't wire it into anything you've already built.
It also can't be self-hosted, which rules it out immediately for any business running on its own infrastructure for data residency or cost reasons. If you're in the GCC, note that Pro-tier dots is excluded in the EEA, Switzerland, and the UK at launch, so plan eligibility varies more by region than most AI launches.
The Honest Trade-Offs#
What's genuinely good here: the permission model. OpenAI built four explicit levels into Custom Rules, from "execute without asking" down to "ask before every action," plus a Pause button and an activity log split into Active, Scheduled, and Completed. That's a cleaner control surface than most n8n flows ship with by default, and I'll say that plainly even though it's not our product.
What's missing is just as concrete:
- No public API or developer access at launch
- No self-hosted deployment option, at all
- Resetting a dot deletes conversations, memory, and scheduled tasks together, with no selective rollback
- The dot can research but can't message, edit, or act on your behalf without a rule allowing it
- Text messaging is a US-only Pro beta; a dot can't call you back at launch
Disconnecting a connected app doesn't erase what the dot already learned from it either. That's a real gap if you ever need to revoke access cleanly.
How We're Already Handling This#
We looked at dots the week it launched and made a call fast: we're not moving a single client workflow onto it, because "no API" is a dealbreaker for anything that needs to trigger off a CRM event or a form submission, not a chat window. Our own builds stay on n8n running on our own VPS stack, with PM2 and Nginx doing the boring, reliable parts.
What we are doing at WebEpex is running a dot internally on our own ops, since late September 2026, for exactly the kind of task it's actually built for: background research and drafting that doesn't need to touch a client system. One of us has it watching a competitor's pricing page and summarizing changes weekly. That's it. That's the whole use case right now, and I'd rather say that plainly than oversell it.
We also noted the overlap with Decagon's Personal Agent Gateway, which name-checks dots as one of the personal agents it's built to detect. Between a product built to spot agents like dots and a dot that can't yet act on a client's behalf, the practical gap for SMB automation is still wide open, and it's the same gap we've been closing with plain self-hosted n8n and scoped access rules, not with anyone's agent platform.
What I'd Tell a Client This Week#
Don't replace anything. If you're already paying for ChatGPT Business Premium, turn dots on and point it at research you'd otherwise assign to a junior hire: competitor tracking, meeting prep, drafting first passes. Don't connect it to anything you can't afford to have learn your data permanently, since disconnecting doesn't erase what it already saw. And if someone pitches you "dots-powered automation" as a replacement for your existing workflow build, ask them for the API docs. There aren't any yet.
If you want a second opinion on whether something like this actually fits your stack, send me what you're running and I'll tell you straight, no charge for the look. cal.com/webepex/growth-review
