FuturPulse analysis · 2 October 2026
OpenAI began rolling out Dots on 29 September 2026, while CopilotKit’s OpenDots remains a self-hosted template: both support text, calls and Slack, but only OpenAI sells a managed always-on service.
At a glance
- OpenAI says Dots run on GPT-6 Astra, have a cloud computer and can connect with more than 4,000 apps.
- OpenDots says each agent can have a separate computer with persistent browser profiles and workspace files.
- CopilotKit Channels supports Slack and Microsoft Teams through one agent process rather than a second channel-only bot.
- OpenDots recorded local verification of calls on 30 September 2026, while Slack still needed connected-service verification.
What shipped, and what remains promised?
OpenAI Dots is a managed product rollout. CopilotKit OpenDots is code that an organization must deploy, configure and operate. That difference matters more than the shared “always-on coworker” label.
OpenAI announced Dots for ChatGPT Pro and Business Premium users in eligible markets, with an Enterprise beta enabled by workspace administrators. It promised a primary Dot, a cloud computer, browser work, Slack and Teams messaging, voice calls, and cross-channel context.
CopilotKit recorded live verification of OpenDots Intelligence, model responses and page-context chat. Its repository also labels OpenDots an early-development template, not a hosted product.
OpenDots recorded local verification of realtime speech, call controls and saved call receipts. The same project says Slack and spoken delegation to a compute agent still need verification against connected services.
OpenAI’s first Dot is included with qualifying plans, according to The Next Web, but regional availability differs by plan. The outlet says Pro users in the European Economic Area, Switzerland and the United Kingdom are excluded for now.
The practical reading is simple. Dots is the choice for a team that wants OpenAI to run the agent service. OpenDots is the choice for a team that wants the architecture, interface and deployment decisions under its own control.
Which product is actually always on?
OpenAI markets Dots as continuously available agents that can pursue work toward a user’s goals around the clock. Its background mode, called proactive research, uses read-only tools in connected apps when the user is away.
OpenAI says Dots can work 24/7, but its unattended mode cannot send messages, change content, or control a browser or computer. That restriction is important: “always on” does not mean unrestricted automation while nobody watches.
OpenDots takes a different route. The project includes scheduled server-side turns, pause controls and retries for background work. Yet a working deployment still needs the operator’s model provider, conversation service, browser containers and credentials.
OpenDots describes background work as scheduled server-side turns in the original conversation. That is a deployable building block, rather than CopilotKit’s promise to run an agent for every user without operational work.
| Decision | OpenAI Dots | CopilotKit OpenDots | What it means for a team |
|---|---|---|---|
| How it is delivered | Managed ChatGPT feature | Self-hosted open-source template | Choose Dots for speed; choose OpenDots for infrastructure control. |
| Computer | OpenAI cloud computer | Per-Dot container-backed computer | OpenDots exposes more deployment choices, but needs setup. |
| Slack | Messaging in Slack is part of the rollout | Channels SDK connection with allowlists | OpenDots gives builders explicit identity mapping and controls. |
| Calls | Voice calls to a Dot | Realtime speech plus a separate compute agent | OpenDots splits conversation from longer-running work. |
| Approval model | Custom Rules and auto-review | Human-in-the-loop cards | Both put a person before consequential work, differently. |
| Model choice | GPT-6 Astra | OpenAI-compatible provider configured by operator | OpenDots can fit an existing model arrangement. |
Comparison compiled from OpenAI’s Dots announcement and the OpenDots repository.
How do Slack workflows differ?
OpenAI Dots treats Slack as another place to reach the same assistant. A person can start work in ChatGPT, share context in Slack, and continue with the Dot there. OpenAI says context follows the Dot across its supported channels.
OpenAI lists Slack and Microsoft Teams as messaging surfaces for Dots. The company also says texting is coming later, so buyers should not treat every communication channel as available today.
OpenDots treats Slack as an integration that a builder enables. A user mentions a selected Dot in a Slack thread, then the managed channel connection forwards the turn to the agent that the organization runs.
CopilotKit Channels says the same agent process serves the application, Slack and Teams. Its replies can use native Slack Block Kit or Teams Adaptive Cards, rather than forcing users into a web-chat link.
This structure gives an internal platform team a useful control point. Its code maps a Slack or Teams person to an application user, so existing permissions can follow that person. The operator decides when memory can be used.
CopilotKit Intelligence can be cloud-hosted or self-hosted in Kubernetes or an AWS account. The documented portability helps teams that want to start managed, then move the persistence layer into their own environment.
What changes when calls are involved?
OpenAI presents calls as a direct way to talk through work with a Dot. The agent can be messaged or called through ChatGPT on desktop, web and mobile, while Slack and Teams remain message surfaces.
OpenAI says users can call a Dot and receive progress, questions or decisions that need attention. The announcement does not publish a technical call architecture, latency target, call-length limit, or list of voice providers.
OpenDots is more specific about the design. It pairs realtime speech with a separate compute agent, allowing a longer task to keep running while the spoken conversation continues. Both parts share conversation context and tool permissions.
OpenDots lists captions, mute controls, a minimized chat view and timeline receipts for calls. Its voice service needs separate provider configuration, which makes it a flexible starting point but not a turnkey phone colleague.
For a sales or support workflow, OpenAI’s route is simpler if the user accepts its service boundaries. For a product team building its own voice agent, OpenDots offers visible components to alter, including tool permissions and the computer panel.
What are ChatGPT dots?
ChatGPT Dots are OpenAI’s persistent AI agents, introduced on 29 September 2026. Each Dot uses GPT-6 Astra and has its own cloud computer and browser, according to OpenAI. A user can name the primary Dot, connect approved apps and ask it to carry work across ChatGPT, Slack and Teams.
They are not ordinary chat threads. OpenAI says a Dot can work on several projects and learn from feedback over time. A Dot can also work proactively with read-only app access when the user is not active. Actions that could affect accounts or share information are subject to rules and review.
What is dots in AI?
“Dots” is a reused name, not a general technical standard. In this comparison, it can mean OpenAI Dots, a managed agent product, or CopilotKit OpenDots, a self-hosted template for building persistent agent workspaces. Both use the word for agents that retain work context and can operate across several interfaces.
There is also another project with the same name. feder-cr/dots describes itself as an AI agent with a browser and says it can use models through OpenRouter. Its focus is browser behavior and local operation, not Slack coworkers, enterprise calls or a shared work-document interface.
The safest procurement question is therefore not “Which Dots?” It is “Who runs the computers, identity, model access and message delivery?” Those answers sharply separate the three projects.
Is AI available on 24/7?
AI software can be available 24/7, but its ability to act without a person depends on its permissions. OpenAI says Dots are available 24/7 and can perform proactive research in the background. That background work is read-only, which limits the chance of an unattended action changing an account or sending a message.
OpenDots can be designed for continuous operation because it includes background turns and persistent computers. However, AG-UI describes long-running agents as software that streams work across multi-turn sessions, not as a guarantee that any particular deployment will run without interruption.
For a business, 24/7 availability should mean monitored service, clear escalation paths and explicit approval limits. It should not mean granting a bot permanent authority to make commitments.
What we could not verify?
OpenAI has not publicly listed pricing for additional Dots, faster Dot capacity, or monthly work capacity. It also has not published the commercial terms for specialist Dots, which it describes as enterprise pilots. OpenAI could settle those points with plan documentation and a specialist-Dot price list.
CopilotKit has not publicly confirmed live Slack verification for OpenDots, nor connected-service verification for spoken delegation. The project itself identifies those gaps. A production deployment report from CopilotKit or a customer operator could settle them.
Neither product page publishes a comparative uptime commitment for an always-on coworker. Buyers who need one should request a service-level agreement, incident process and data-retention terms before giving an agent access to sensitive systems.
Sources
- https://openai.com/index/introducing-dots/
- https://github.com/CopilotKit/OpenDots
- https://docs.copilotkit.ai/intelligence/channels
- https://docs.copilotkit.ai/intelligence/overview
- https://docs.ag-ui.com/introduction
- https://github.com/feder-cr/dots
- https://thenextweb.com/news/openai-dots-always-on-ai-agents-cloud-computers-devday
