AI NewsOpen sourceAnnouncement
CopilotKit released OpenDots, a self-hosted template for always-on agents, and it passed 1,295 stars in three days
CopilotKit published OpenDots on 29 September 2026, a self-hostable template for always-on AI agents with Spaces, specialist roles and per-agent computers, and the repository has reached 1,295 stars in the three days since.
Image: GitHub
Why it mattersA team that wants the shape of OpenAI's new Dots product without sending its data through OpenAI now has an MIT template with the same specialist agents, per-agent computers and Slack bridge, running against any OpenAI compatible provider.
OpenAI launched Dots on 29 September, giving each user a set of named agents that work in text, calls and Slack. CopilotKit put a self-hostable template with the same shape on GitHub the same day.
The repository, CopilotKit/OpenDots, is MIT licensed and had 1,295 stars and 155 forks by 2 October 2026, three days after creation. It is marked alpha and described in its README as "a starting point for building your own agent workspace. Clone it, define your Dots, connect your services, and adapt the interface and tools to your needs."
What is in the template
A Space is a document workspace with a searchable library, nested pages, and a visual editor with slash commands. Each Dot is a specialist agent with a name, role, instructions and permitted tools, and can hold access to several Spaces. Pages save to a local database with autosave, revision checks and markdown source mode. A researcher Dot can investigate a topic and a writer Dot can turn the findings into a draft, with the saved page linking the two conversations.
Each Dot can run its own computer, built on CopilotKit's earlier OpenBot container supervisor. The Computer panel exposes browser control, human takeover, files, terminal output and activity, with browser, file and shell permissions set per Dot. A different computer credential is derived for every Dot, so one Dot's work cannot reach another's state. The template also includes a human in the loop card, so a Dot can be asked to show a draft and wait for an Approve and save decision before the page is created.
Text, calls and Slack
A single conversation keeps text and call controls together above the Dot's avatar. Calls pair real time speech with a separate compute agent, so a long running task continues while the user is on the line. The call screen has a live timer, separate user and Dot captions, microphone and speaker mute, and a minimised view. Voice needs a separate provider configuration.
Slack integration goes through CopilotKit's Channels SDK and OpenTag, with an explicit workspace and user allowlist and a chosen specialist Dot. Mentioning the Dot in Slack opens a thread in its own context.
Where this sits against OpenAI Dots
OpenAI Dots is a hosted product that keeps the user's data, the agent configuration and the computer environment inside OpenAI's account. OpenDots gives up the hosting and takes on the setup in exchange for keeping all three on the operator's own infrastructure, with any OpenAI compatible provider under the model layer. The template uses TanStack AI for streaming, AG-UI to carry messages between the backend and the React components, and stores pages, Dots and conversations in a local workspace database.
A team that was planning to adopt Dots inside a company with a data policy that forbids sending customer material to OpenAI now has a worked example to run against its own stack. The template warns that it is in early development and that an unconfigured copy will not execute commands on the host.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
