OpenAI Dots and Meta Muse are always-on AI agents that work for one account holder. Dots come with ChatGPT Pro and Business Premium; Muse is free, with $20 and $100 plans. Buy one for a person's admin and research. Build a custom agent when customers use it, regulated data is involved, or it writes to your system of record.
Updated 13 November 2026. Since launch, OpenAI has moved dots for Enterprise, Edu and Healthcare from beta to an admin preview that ends on 18 November, when billing starts and dots switch on by default unless an admin opts out. Muse has added Singapore. For a team of five, dots cost about $500–$625 a month in Business Premium seats. A custom agent costs $15,000–$75,000 to build before it saves anything. Once the agent runs inside your product at 100,000 tasks a month, neither off-the-shelf option fits. For agent build costs in detail, see how much an AI agent costs. If you have already decided to build, see our guide to AI agent frameworks in 2026.
OpenAI Dots are always-on agents inside ChatGPT that keep working on a goal after you close the chat. They are for individual professionals and employees who want an assistant to run their own errands across the apps they already use.
OpenAI announced dots on 29 September 2026. Each dot has its own cloud computer, reaches more than 4,000 apps through ChatGPT plugins, and can be messaged in ChatGPT, Slack or Microsoft Teams. The launch post was updated on 3 November to say dots use GPT-6 Astra and GPT-6.1 Terra for different parts of a task. Availability as of 13 November 2026:
The most sensitive actions, such as changing a password or moving money, still require you to take over and finish them yourself.
Meta Muse is a personal AI agent from Meta that acts on your behalf in apps such as email, Instagram and online shops. Businesses can use it through Muse for Small Business, which is built for owners of small companies in the US and Canada, not for teams with central admin controls.
Muse launched on 8 September 2026 on iOS, Android, the web at muse.ai and WhatsApp, and reached Singapore in October. Meta says Muse does not share your conversations or the data in its virtual machine with Meta's ad systems, and that you can opt out of model training. Muse for Small Business followed on 29 September. It connects to Asana, Box, Canva, Dropbox, Figma, Granola, HighLevel, Intuit QuickBooks, Klaviyo, Lovable, Notion, Shopify, Slack, Stripe and Zoom, plus Facebook Pages, Instagram professional accounts and Meta ad accounts. Meta states that nothing publishes, sends or spends without your approval. The announcement describes no admin console, SSO or compliance commitments.
Neither product is something you embed in your own app for your own customers. Both act for the account holder. That one fact settles most build-versus-buy questions before cost comes into it.
Your first dot is included in a ChatGPT Pro plan ($100, $200 or $500 a month) or a Business Premium seat ($125 a month, or $100 billed annually). Muse is free for most use, with Power ($20 a month) and Maximum ($100 a month) plans. Extra dots have no published price yet.
| Plan | Price (13 Nov 2026) | What you get |
|---|---|---|
| ChatGPT Pro | $100, $200 or $500 a month | First dot included; not available in the EEA, Switzerland or the UK |
| ChatGPT Business Premium | $125 per user a month, $100 billed annually | First dot included per Premium seat |
| ChatGPT Enterprise, Edu, Healthcare | Contract pricing: flat rates by type of work, or consumption | Free preview to 18 Nov 2026; billed from 18 Nov. Work handed to ChatGPT Work or Codex is billed separately |
| Muse (free) | $0 | Meta says it covers "most of what people need" |
| Muse Power / Maximum | $20 / $100 a month | Larger weekly usage allowance; extra tokens can be bought |
OpenAI says you will later be able to add more dots and buy more speed or monthly capacity per dot, but those prices are not public. Meta's small-business announcement does not list separate business pricing, so assume the consumer Muse plans apply until Meta says otherwise.
There are four realistic paths. "DIY on a vendor platform" means building your own agent on hosted agent infrastructure, such as OpenAI's Agents API (public beta since September 2026, with computer use, hosted execution and multi-agent support) or Claude Managed Agents (billed as tokens plus $0.08 per session-hour). "Custom" means an agent whose code you own, built by your team or a studio.
| Criterion | OpenAI Dots | Meta Muse for Small Business | DIY on a vendor platform | Custom agent |
|---|---|---|---|---|
| Who it works for | An employee, inside ChatGPT, Slack or Teams | A small-business owner (US and Canada) | Your users, inside your product | Your users and your systems |
| Time to first value | Minutes | Minutes | Weeks | 6–12 weeks for a single agent |
| Price model | Per seat on Pro and Business Premium; contract billing on Enterprise from 18 Nov 2026; extra dots not yet priced | Per person: free, $20 or $100 a month | Tokens plus platform fees plus your engineering time | Build cost plus tokens |
| Regulated data (PHI, card data) | PHI allowed only in an eligible workspace under a BAA, with dots enabled for your role; connected third-party apps keep their own terms | No compliance commitments in the launch materials | Possible where the vendor signs a BAA for that API | Designed in: masking, scoped access, audit logs |
| Fixed, repeatable process | Learns preferences; not a workflow engine | Goal-driven; not a workflow engine | You code the process | You code the process; the model proposes, code validates |
| Writes to your system of record | Through plugins and their permissions | Through its connectors | Through tools you write | Through tools you write, with idempotency and limits |
| Approvals | Default rules plus Custom Rules; password changes and money transfers need you to take over | Nothing publishes, sends or spends without approval | Whatever you build | Whatever your risk policy requires |
| Lock-in | High: tied to a ChatGPT plan | High: tied to a Meta account | Medium: tied to a vendor SDK and runtime | Low, if the code is yours and the model can be swapped |
The lock-in row is not hypothetical. OpenAI's deprecations page shows Agent Builder, the visual builder announced with AgentKit, was deprecated on 3 June 2026 and shuts down on 30 November 2026. Teams that built on it now have to move to the Agents SDK or ChatGPT workspace agents. Any hosted agent product can change under you on the vendor's schedule.

For a company with staff, IT and Slack or Teams, Dots is the better fit: Business Premium and Enterprise give admins control over who uses dots and what they can reach. Muse fits an owner-operator whose business runs on Shopify, QuickBooks, Stripe and Meta ads, and who approves everything personally.
More often than agencies like to admit. Our own AI agent development page says the most useful thing an agent builder can tell you is when you don't need one. Buy first if most of these are true:
A custom agent wins when the agent is part of your product or your operations rather than one employee's assistant. In practice, one of these is usually true:
There is a middle path too. If your goal is for other people's dots to use your product, you don't need an agent. You need a clean API or an MCP server that agents can call. We cover that in MCP server development.
Yes, within limits. A dot reaches your systems through ChatGPT plugins, including a plugin you build around your own MCP server; Muse reaches them through Meta's listed connectors or a custom connector. Either way the agent still works for one account holder, so your system sees one user's permissions, not a service of your own.
The table compares monthly running cost plus year-one cost in three scenarios. Seat prices are OpenAI's and Meta's published prices on 13 November 2026. Custom-agent token costs use the assumptions in our AI app cost guide: 60,000 input and 3,000 output tokens per agent task, on a model priced at $2 / $10 per million tokens (GPT-6 Sol), with no caching. The custom build range of $15,000–$75,000 is the one we publish for a scoped single-purpose agent.
| Scenario | Dots (Business Premium) | Muse for Small Business | Custom agent | Likely winner |
|---|---|---|---|---|
| 5 staff, ~1,000 tasks/month, internal admin | $500–$625/mo, about $6K–$7.5K in year one | $0–$500/mo, depending on plan | $15K–$75K build plus ~$150/mo tokens: about $17K–$77K in year one | Buy |
| 50 staff, ~10,000 tasks/month, shared process | $5,000–$6,250/mo, about $60K–$75K in year one | Not built for a 50-person shared process | Build plus ~$1,500/mo tokens: about $33K–$93K in year one, tokens only from year two | Close in year one; custom from year two if the process is stable |
| Customer-facing, ~100,000 tasks/month | Not available: dots act for the account holder | Not available | ~$15,000/mo on the mid tier; ~$750/mo on GPT-6 Luna ($0.10 / $0.50 per million tokens) if it passes your evals | Custom |
Treat these as orders of magnitude, not quotes. OpenAI hasn't published a dot's monthly capacity beyond the included allowance, so a busy dot may need extra capacity that isn't priced yet. On Enterprise, dots are billed under your contract from 18 November. Custom token costs fall sharply with routing and caching. And the build range moves with integrations and compliance. Our MVP cost page puts compliance at +15–25% when it is designed in and +40–80% when it is bolted on later.

Run this checklist before you give a dot or Muse access to company systems:
Ask four questions in order. Stop at the first "yes".
If none of these fit cleanly, the honest answer may be neither. A workflow with fixed inputs and fixed rules is often better served by ordinary software than by any agent, and it will be cheaper, faster and more reliable.

For one person's own work across mainstream apps, often yes. For anything your customers use, anything that must follow a fixed process, or anything that touches regulated data without a covering agreement, no. A dot acts for the ChatGPT account holder and is not something you embed in your product.
Meta says nothing publishes, sends or spends without your approval, that checkout uses one-time card numbers through Link by Stripe, and that conversations are not shared with Meta's ad systems. Those are good defaults. Still connect only the accounts it needs, review every payment it proposes, and keep your own records.
Muse's launch materials make no HIPAA commitment. OpenAI says dots can be used with PHI from an eligible ChatGPT workspace under a BAA, when dots are enabled for your role; other features and third-party apps a dot uses keep their own coverage. Confirm your agreement and each connected app before any PHI goes near a dot.
A hosted agent platform gets you running faster but ties you to that vendor's runtime and roadmap. OpenAI's Agent Builder is shutting down on 30 November 2026, which shows the risk. Keep your business logic, tools and evals in your own code, so the hosted pieces can be swapped.
We build agents whose code you own, on a design you can inspect:
BeevR is a senior, founder-led AI studio in Hanoi, Vietnam. We build custom AI agents at a fixed price per phase, you own the code and IP from day one, and we build for regulated industries. If you're weighing a dot against a custom build, tell us about the workflow. We'll tell you honestly whether you need us.