Grok Bot vs. ChatGPT Dots vs. OpenDots: A Practical Guide

If you have been following AI agents, you have probably seen Grok Bot, ChatGPT Dots, and OpenDots mentioned together. They all point toward assistants that can keep working beyond a single exchange. Choosing between them, however, means looking past the similar names.

Grok Bot and OpenAI Dots are vendor offerings. CopilotKit's OpenDots is a template for building and running your own agent workspace. There is also a separate OpenDots project from Digger. That difference affects who maintains the system, where its data goes, and how much setup your team takes on.

This comparison draws on official announcements and maintainer documentation checked on October 9, 2026. It is a guide to evaluating the options, not a hands-on performance benchmark.

Cover: an original AI-generated illustration for Makezaa. The three illustrated assistants are concepts, not vendor logos or product screenshots.

Grok Bot vs. ChatGPT Dots vs. OpenDots at a glance

Option	What the source describes	What to check first
Grok Bot	Persistent agents with their own computers, roles, and routines	Current access, connected tools, permission controls, and routine behavior
OpenAI Dots	Always-on agents available through ChatGPT and supported work channels	Plan and market eligibility, app access, and action review settings
CopilotKit OpenDots	An open-source template for a custom agent workspace	Intelligence backend, model configuration, hosting, and live integration status
Digger OpenDots	A separate, model-agnostic personal agent built on OpenComputer Serverless Agents	The exact repository, runtime dependencies, and deployment requirements

Grok Bot: ongoing roles, computers, and routines

In its September 3, 2026 design article, x.ai explains how Grok Bot supports agents that persist across sessions. The interface includes a roster of Bots, visible activity, and each Bot's own computer. Routines can activate work on a schedule or in response to an event.

The article also separates shared capabilities from role-specific context: tools and Skills sit at the account level, while memory and routines belong to individual Bots. That is a useful distinction when several assistants need the same tools but different responsibilities.

For a business, the practical test is whether those responsibilities stay understandable. Can you tell which Bot owns an assignment? Can you review a routine's output? Can you stop the work or help when it gets stuck? Use the current product and documentation to answer these questions; a design article alone does not establish pricing or availability for your account.

ChatGPT Dots: OpenAI's approach to ongoing assistance

OpenAI announced Dots on September 29, 2026. Its description includes GPT-6 Astra, individual cloud computers, feedback-based learning, and access through ChatGPT, Slack, and Teams. The announcement says rollout is beginning across Pro, Business Premium, and Enterprise plans in eligible markets.

OpenAI distinguishes proactive research using read-only connected-app tools from actions taken during assigned tasks. It describes controls including Activity View and Custom Rules. The term “always-on” should therefore not be read as unrestricted permission to take every action.

Before adopting Dots for a team, check what your account can actually use. Then run a bounded assignment with clear completion criteria. A research task might end with a source-backed brief; a publishing task should end with a verified page and a record of what changed.

CopilotKit OpenDots: a workspace you configure

CopilotKit/OpenDots is an independent open-source template, not an official edition of OpenAI Dots. Its workspace includes Spaces, specialist Dots, computer tools, and agent interfaces.

Conversations require CopilotKit Intelligence. The repository lists hosted Intelligence, local Docker evaluation, and licensed self-hosted deployment options. You also configure a model provider and the integrations you need. An MIT-licensed template does not mean every connected service, model call, or deployment is free.

The maintainers describe the template as early development and distinguish local tests from connected-service verification. Read those notes before assuming that a demonstrated interface proves every integration is ready for your use case.

This route gives a development team room to adapt the experience. It also gives that team operational work: upgrades, permissions, integration failures, data retention, and deployment maintenance. Include that work in the decision, rather than comparing only the subscription price of a hosted assistant with the license of a repository.

Digger's OpenDots is a different project

diggerhq/opendots describes a model-agnostic personal agent built on OpenComputer Serverless Agents. It is separate from CopilotKit's template. When an article or demo says “OpenDots,” check the repository and maintainer before following its setup instructions.

Five questions that make the comparison useful

1. Who will maintain the agent?

A hosted offering can reduce the infrastructure work your team owns. A custom deployment may offer more control, but someone must monitor it and fix failed integrations. Decide who that person or team is before committing to either model.

2. What happens to your data?

Map the whole path: the workspace, conversation backend, model provider, connected apps, and browser environment. Self-hosting one component does not establish that the entire workflow stays on your infrastructure. Review the services you actually configure.

3. What actions can it take?

Separate reading, editing, publishing, sending messages, and deleting. For an editorial workflow, you may want research and drafting to run freely while publication follows an explicit instruction. Choose controls that fit the task rather than treating autonomy as a single switch.

4. What happens after an interruption?

Try a small recovery test. If the agent saves a draft and then loses its connection, can it resume without creating another copy? Does the report distinguish a saved draft from a published page? Recovery behavior is worth checking before trusting a recurring assignment.

5. Can you verify the result?

Ask for evidence appropriate to the task: source links for research, a preview for a design, tests for a code change, or a working public URL for publication. A confident final message is easier to trust when you can inspect the work behind it.

Choose around the work you want to delegate

If your priority is getting started with limited infrastructure work, evaluate a hosted option against one real task. If you need a custom workspace and have people to maintain it, investigate OpenDots with its backend and integration requirements in view.

There is no useful universal winner without a defined workload. Run the same brief through the options available to you, record corrections and setup effort, and compare the results your team can accept.

For a concrete brief and publishing checklist, read our practical AI agents guide. If the missing piece is connecting an agent to your website, talk to Makezaa about your workflow.