CopilotKit founders pitch self-hosted OpenBot directly against Grok Bot

CopilotKit co-founders Atai Barkai and Uli Barkai published a three-post thread on X on September 22nd, 2026, showing their self-hosted OpenBot agent team and SpaceXAI's managed Grok Bot service running the same four-agent travel-planning assignment with the same prompt, positioning OpenBot as the open-source alternative to managed AI coworkers. OpenBot, CopilotKit's open-source platform for deploying persistent AI coworkers inside an organization's own infrastructure, shipped its v0.0.14 release on September 21st, 2026, and CopilotKit said the project passed 3,500 GitHub stars by September 1st. The Seattle-based company raised a $27M Series A on May 5th co-led by Glilot Capital, NFX and SignalFire to support its enterprise agent interface stack and AG-UI work.

CopilotKit founders pitch self-hosted OpenBot directly against Grok Bot Atai and Uli Barkai are using a four-agent travel demo to frame OpenBot as the open-source alternative to SpaceXAI's managed agent service, even if the clip is no benchmark. By Ryan Merket https://runtimewire.com/author/ryan-merket · Published Primary source: X https://x.com/OpenBot /status/2102405058415493356 Why it matters OpenBot is using Grok Bot's launch to frame self-hosting as the alternative to managed AI coworkers, but the first comparison offers positioning rather than reproducible performance evidence. CopilotKit co-founders Atai Barkai @ataiiam https://x.com/ataiiam and Uli Barkai are giving their open-source agent project a conspicuous rival: Grok Bot. In a three-post thread on X https://x.com/OpenBot /status/2102405058415493356 published September 22nd, 2026, the founders' OpenBot project showed its self-hosted agent team and SpaceXAI's managed service attempting the same travel-planning assignment. The comparison is less a declaration of a measured winner than a statement about where the Barkais intend to compete. Each setup included four named specialists: a Traveler Assistant, Flight Bot, Accommodation Bot and Attractions Guide. OpenBot said the teams received the exact same prompt and that both runs were recorded. The video puts OpenBot beside a higher-profile product without pretending the two share the same operating model. Grok Bot supplies the models, cloud computers and interface as a hosted service. OpenBot gives organizations source code and infrastructure controls, then leaves deployment, model selection and operation to them. That distinction is the product thesis the Barkais are selling. Companies interested in persistent AI coworkers should not necessarily have to rent the entire stack from one provider, they argue. They should be able to choose the models, agent frameworks, tools and policies themselves, then run the system inside infrastructure they control. The founder bet behind OpenBot OpenBot https://www.copilotkit.ai/openbot is CopilotKit's open-source platform for deploying persistent AI coworkers inside an organization's own infrastructure. Administrators can assign roles, connect company documents and tools, and apply policy checks to browser, file and MCP actions. The platform does not bundle a model; operators provide model credentials and choose how each agent is implemented. The software extends the strategy Atai and Uli Barkai have pursued through CopilotKit: agents should work inside applications, share tasks with people and remain portable across model and framework choices. OpenBot connects agents through AG-UI, the open interaction protocol maintained by CopilotKit. Teams can bring agents built with frameworks including LangGraph, Mastra, CrewAI and Pydantic AI. Each AI coworker can receive its own browser, files and workspace, while actions pass through configurable policy and audit controls. The OpenBot repository https://github.com/CopilotKit/OpenBot describes the software as an alpha template that organizations clone and operate themselves, not a finished hosted subscription. Its listed v0.0.14 release shipped September 21st, 2026. CopilotKit said the project had passed 3,500 GitHub stars by September 1st, a company-reported milestone followed by a rapid series of releases. The project also gives the Barkais a product layer above CopilotKit's developer components. Their Seattle-based company raised a $27M Series A on May 5th https://www.copilotkit.ai/blog/series-a , co-led by Glilot Capital, NFX and SignalFire. CopilotKit said the financing would support its enterprise agent interface stack and AG-UI work. OpenBot turns that infrastructure thesis into something buyers can evaluate as a workplace product: named agents, persistent responsibilities, individual workspaces and controls for the actions those agents take. Why the founders picked Grok Bot Grok Bot launched on August 11th https://x.ai/news/introducing-grok-bot as a hosted service for persistent agents that work on cloud computers, sign into applications and coordinate with other bots. SpaceXAI later expanded access across SuperGrok and Cursor subscriptions, then opened Grok Bot to enterprise customers https://x.ai/news/grok-bot-for-enterprise on September 3rd with access, network and audit controls. That expansion moved Grok Bot into the category CopilotKit wants OpenBot to occupy: persistent, tool-using workers that can be assigned specialized roles and coordinated as a team. The travel-planning clip lets the Barkais argue that the experience does not belong exclusively to a managed platform. OpenBot can also arrange four specialized agents around one assignment, while offering source access and infrastructure control. For founders trying to turn an open-source developer stack into an enterprise product, that is the strategically useful comparison. It is not, however, a reproducible performance test. The posts did not publish the full prompt, underlying models, tool permissions, browser state, starting context, completion times, costs, raw traces or a scoring rubric. They also did not identify a measured winner. Those omissions matter because matching agent names and an initial prompt controls only a small part of an agent run. Model selection, search access, stored memory, browser permissions and the instructions assigned to each specialist can all change the result. The demonstration therefore supports a narrower claim than a head-to-head victory: the Barkais have built an open-source, self-hosted agent team capable of attempting the same class of multi-agent work that SpaceXAI packages as a hosted product. For OpenBot, that may be the more important message. The founders are not merely showing that four agents can plan a trip. They are positioning control of the agent stack - models, infrastructure, frameworks and policies - as the reason companies might choose OpenBot over Grok Bot.