When OpenAI introduced dots last month, the company described them as "always-on agents" operating within their own cloud computer, but numerous users are now reporting widespread technical failures and connectivity problems, according to a report published by The Register on October 7, 2026. The difficulties include task breakdowns, cloud connection errors, dots that vanish entirely, and rejections that seem tied to abuse prevention systems, with similar complaints appearing across OpenAI's Developer Community forum.
The technical stumbles haven't stopped competitors from launching their own alternatives. A firm called Tawkit released OpenDots, a fully open source version that users can host themselves, allowing any OpenAI-compatible model to run through their AG-UI agent layer. Another project called Open Dot offers a potentially cheaper option powered by AI models running on OpenRouter, with plans to support local model execution. Created by Prathit Joshi, a growth engineer at Composio based in Bangalore, India, Open Dot runs as a desktop app built on the Electron framework and connects to services like Gmail, Calendar, Slack, Notion, and GitHub through Composio.
According to the report, ABI Research argues that "AI claws" – always-on agents – will shape enterprise AI adoption, describing "a market evolving from isolated assistants toward modular agent ecosystems built for longer-running, enterprise-grade automation." Joshi told The Register he built Open Dot because he enjoys "building and shipping tools for developers, especially tools that can support use cases that closed model companies aren't necessarily incentivized to support, like allowing open source models." He's been using the tool to manage personal tasks such as creating daily to-do lists from Slack messages and summarizing social media posts.
The proliferation of dots alternatives reflects a broader pattern in which companies wrap increasingly commoditized AI models in proprietary interfaces, hoping to build sustainable businesses. The open source versions don't eliminate the token costs of whatever agent gets deployed, but they do remove OpenAI from the billing relationship and potentially offer lower-cost options. On Tuesday, Elon Musk announced that Grok Bot would shift from a front-end for a specific model to a model router, saying "SpaceX will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs." Despite dots drawing both enthusiastic users and detractors who question their purpose, along with speculation that OpenAI rushed the launch in response to Meta Muse and Grok Bot, the rush to create competing always-on agent platforms suggests companies see persistent agents as the next battleground for AI adoption. The challenge for all of them will be proving that businesses and consumers actually want agents running constantly in the background, rather than assistants they summon on demand.