OpenAI Dot: Persistent Agent, Restricted Access, and Concrete Trials

OpenAI deploys Dot, an agent that remains active between exchanges and can work on a dedicated cloud computer. Access is governed by eligibility criteria and technical limits, while trials on analytical and planning tasks demonstrate structured execution and effective programming. It remains to be seen whether this persistence will produce genuinely useful work at scale.

Usage Limits: Quotas, Integrations, and Enterprise Constraints

Interactions with Dot do not count against the usage limits of ChatGPT, but the Work or Codex tasks it initiates or manages are charged to the quotas of those products. A plan includes enhanced work allocation, with extended limits for the first month after launch, though no figures have been published for Dot. In terms of access, Slack and Teams can serve as interfaces to the same agent, subject to availability; adding Dot to Slack does not trigger channel monitoring, and Teams remains in alpha by invitation with messaging expected soon. For enterprises, the beta excludes data residency or inferences and does not support FedRAMP, EKM, and AE/UAE inference workspaces. Local access to a company computer requires the application version 26.929 or newer, though this does not apply to creating a Dot cloud. By default, Dot is disabled and must be activated by an administrator, in a phased deployment where eligibility does not guarantee immediate access.

Access: Eligibility, Excluded Areas, and Administrator Activation

Dot is offered globally to Pro 100, Pro 200, Pro 500 subscribers, and Business Premium plans, with an age requirement of over 18 years outside the EEA, UK, and Switzerland. Free, Go, Plus, standard Business, and Edu accounts are not eligible. An interface message may indicate a restriction related to the account and region without invalidating an entire country; in this case, it is advisable to check the current eligibility rules and, for business accounts, the administrator settings.

Workstation: 24/7 Cloud or Single Local Computer

Dot can be created from the desktop application of ChatGPT or a browser on a computer, via a dedicated entry that presents its purpose as well as links to data controls. Application connections may be offered for configuration and added later. Once set up on desktop, Dot is supported on mobile when the corresponding update is available. Dot can work on its cloud computer, selected for 24/7 access, or on a local computer connected via the application. The cloud mode continues tasks even if the user's device is turned off, while the local mode requires an online computer with ChatGPT open and limits the connection to a single personal workstation at a time. A provisioning screen indicates the progress of the setup. The first conversation invites customization of the agent, including name and appearance, and may suggest tasks based on context.

Analytical Execution: Accurate Calculations, Caution on Spending

A trial focused on a synthetic analytical review in Markdown based on fictitious paid research data and rankings, without recurring planning. The datasets included for August and September volumes of leads, customers, and revenue by channel. Deliverables required conversions by channel and overall, revenue growth, and the change in percentage points of overall conversion, a three-sentence recommendation distinguishing available and missing data on spending, profit, and retention, as well as a four-item checklist, all within 350 words. Independent recalculation validated the conversion method (customers grouped on leads grouped), a growth of $3,400 or 13.6%, and a decrease of 6.52 percentage points in overall conversion using unrounded rates (6.53 if subtracting already rounded values). Dot did not endorse the idea of doubling spending solely based on revenue increase, identified missing data, avoided inferring unsupported causality, and proposed a measured budget test. The response adhered exactly to the expected structure and lengths. The whole distinguished a factual indicator from an unproven business conclusion and provided concrete verification points.

Recurring Planning: Successful Scheduling and Activity Tracking

A second scenario requested the scheduling of a weekly check: for the next four Mondays at 10:00 Asia/Kolkata, check a provided reference journal, update a table (model, test, score, date, source), report missing fields or conflicting executions, write a 150-word explanation of significant changes, keep routine updates in the table, and notify in ChatGPT only in case of notable changes or need for arbitration, then confirm schedule and end date, without publishing a draft or contacting others. A reference journal template was provided to frame the format. The interface includes a status tab to track activity and the agent's "thoughts." The task was successful, and a recurring event was created in the scheduling tab. Previous uses of scheduling in ChatGPT were described as slow and unreliable; the experience with Dot was smoother and more intuitive.

Product Goal and Unknowns: An Agent Designed to Last

Dot is a persistent agent powered by GPT-6 Astra, equipped with a cloud computer and a browser, designed to retain context between exchanges and prompt the user when necessary. Its single-session logic aims for continuity and fits into an approach where AI remains in the workflow, uses connected tools, and pursues a goal rather than waiting for each instruction. This orientation brings the tool closer to the idea of a persistent AI collaborator.

However, it remains uncertain whether this persistence will translate into useful value in general. If the agent proves it can reliably handle routines, escalate uncertainty, and know when to involve the user, the practicality of agents could progress. At this stage, it is an early glimpse, illustrated by quick access obtained after launch, and by use cases that confirm certain capabilities without resolving all unknowns.