covey and OpenAI dots
Always-on agents in OpenAI's cloud, and a platform that runs agents on your own servers.
Last checked: 29 September 2026, the day dots were announced, against OpenAI's announcement and its safety article. If something here is out of date, tell us and we will correct it.
dots are OpenAI's always-on agents. Each dot has its own cloud computer and browser, runs on GPT-6 Astra, connects to the apps you allow and can be reached in ChatGPT, by text message, in Slack and in Teams. The first dot is for one person and works on their behalf. OpenAI also previews specialised dots for organisations, with their own identity, credentials and a defined area of responsibility.
covey starts from that second case. It runs agents that belong to an organisation: each has an identity, a container of its own, a manager in the org chart, a backlog and a budget, and it all runs on servers you operate. The engine is an existing coding harness, Claude Code or Codex today.
dots were announced on the day of this check. Everything said here about dots rests on OpenAI's two articles, so every row is marked as a vendor claim.
What it is
| Topic | OpenAI dots | covey |
|---|---|---|
| Kind of product | Always-on agents in ChatGPT. The first dot works for one person; specialised dots for organisations are in a preview with selected companies.Vendor claim[1] | A self-hosted platform that runs an organisation's agents and treats them like staff: a role, a manager, a backlog, a budget. |
| Where it runs | In OpenAI's cloud. Each dot has its own cloud computer; you can also connect your own computer.Vendor claim[1][2] | On your own hosts, one Go binary and Postgres. Each agent gets its own container with a persistent home. |
| Model | GPT-6 Astra.Vendor claim[1] | The engine is Claude Code or Codex, behind a narrow interface. Whoever runs covey chooses the provider. |
| Availability | Rolling out to ChatGPT Pro and Business Premium in eligible markets; a beta for Enterprise workspaces; specialised dots through pilot projects.Vendor claim[1] | AGPL-3.0, installed from GitHub without an account. A hosted version on covey.work is in closed beta. |
Access and control
| Topic | OpenAI dots | covey |
|---|---|---|
| Credentials | Secure sign-in: the model pauses while you enter a password, and stored passwords come from a separate encrypted service without entering the model's context.Vendor claim[2] | Credentials stay encrypted in the platform. A broker gives each run short-lived access scoped to its task; nothing long-lived enters the sandbox. |
| Rules for actions | Built-in rules for when a dot may act, must ask or must hand a step back, plus custom rules. Some actions, such as permanent deletion, always need your confirmation.Vendor claim[1][2] | Rules per action (allow, require approval, deny), set by the organisation. Approvals go to people, and the default rules ship with every installation. |
| Where the check sits | Auto-review, a separate system, checks planned actions before they run. Its controls sit outside the environments dots can change.Vendor claim[2] | Rules are enforced outside the runtime: at the broker, at the egress proxy and in the action layer. |
| Following the work | An activity view in the desktop app shows running and delegated tasks and lets you redirect or stop a dot.Vendor claim[2] | A recording of every run with its tool calls and approvals, token cost per agent and model, and one kill switch for all agents of the organisation. |
Also different
- Whose agent it is
- A dot works on behalf of the person who created it, and learns that person's preferences. A covey agent belongs to the organisation: it has a place in the org chart next to people, a manager who approves, and spend that has an owner.[1]
- Outbound network
- With the docker provider, a covey sandbox reaches only the hosts on its agent's allowlist, by default. OpenAI describes sandboxing of code and tools inside a dot's environment; we found no statement about limiting where a dot's computer may connect, so we make none.[2]
- Your data
- covey runs on your infrastructure; what reaches a language model follows from the tasks you give. OpenAI states that content from Business, Enterprise and Edu workspaces is not used for training by default.[1]
Where dots are ahead
- Nothing to operate: a dot starts inside ChatGPT, with no server, container or update to look after.
- Reach: more than 4,000 apps through OpenAI's plugin ecosystem (vendor figure), and dots answer in ChatGPT, by text, in Slack, in Teams and by voice call.
- Proactive research: dots look for ways to help in the background, with read-only tools.
- A model and an agent built by the same company, with a system card and safety evaluations published alongside.
- Work with Microsoft to govern specialised dots through Agent 365 and Microsoft's identity controls.
Which one fits
OpenAI dots
Choose dots if your people already work in ChatGPT, each of them wants an assistant that acts for them, and running agents in OpenAI's cloud fits your data rules.
covey
Choose covey if the agents are to work for the organisation rather than for one person, run on servers you control, use an engine you pick, and answer to rules, approvals and a recording that you own.
Sources
Checked on September 29, 2026. Statements about OpenAI dots come from its own public sources: documentation, repository, release notes or website. Anything marked as a vendor claim appears only on marketing or pricing pages.
Other comparisons
Paperclip
The closest relative: open source, self-hosted, agents run like staff. Paperclip runs agents as host processes by default; covey gives each one its own container.
Checked on September 29, 2026Governance layerMicrosoft Agent 365
Agent 365 registers and governs agents in Microsoft's cloud. covey is a self-hosted platform where agents do their work.
Checked on September 29, 2026Framework and platformCrewAI
CrewAI is a Python framework for programming agents, plus a platform to deploy them. covey runs agents you describe in configuration.
Checked on September 29, 2026SaaS workforceRelevance AI
Relevance AI builds and runs agent teams in its own cloud, without code. covey runs agents on your hosts, each in its own container.
Checked on September 29, 2026