OpenAI is testing a Codex dashboard that shows which chats are eating your usage limits

RUNTIMEWIRE INVESTIGATION — Scoop

Original reporting by RuntimeWire, based on reverse engineering.

Why it matters

Codex already shows account-wide capacity and reset times. The new attribution would help users identify the specific agentic job, model settings and subagent activity responsible for an unusually large drawdown.

Reporting record

OpenAI is testing a Codex dashboard that traces five-hour and weekly usage back to individual agentic conversations.

The interface ships inside Codex for Windows production build 7868 under a gated Settings route called “Consumer view.” It can rank up to 100 recent Work and Codex chats by their lifetime share of a user’s five-hour allowance, weekly allowance and credit use, according to RuntimeWire’s analysis of the packaged application.

OpenAI already documents a Usage & billing dashboard for eligible Enterprise workspaces. That interface includes “Top consuming chats,” which ranks locally available conversations by lifetime credit use. OpenAI says Personal Analytics and some related features may not be available to every Enterprise and Education workspace.

The implementation in build 7868 adds a different layer of attribution. It compares individual chats against five-hour and weekly allowances and introduces a system designed to associate discovered subagent activity with the conversation that initiated the work.

Account conditions and server-controlled feature gates determine whether Consumer view appears. Internal descriptions identify it as a preview for selected Enterprise and Education members. RuntimeWire could not confirm whether OpenAI has activated the new view for customers.

Top chats adds rate-limit attribution

The new table contains four columns: the local chat title, percentage of the five-hour limit, percentage of the weekly limit and credits used. It sorts by weekly usage by default, though users can select any of the three measurements as the ranking field.

The percentages measure a chat’s lifetime consumption against the user’s current full limits. A long-running conversation can therefore exceed 100 percent after spanning several allowance windows. The measurement attributes historical consumption; it does not show how much capacity remains in the current window.

The table initially displays five rows and can expand when more chats qualify. Opening a row takes the user to the underlying conversation. Additional details can identify the product, model, reasoning effort, speed and funding source behind the usage.

Funding labels distinguish plan usage, credits, a combination of the two and tasks that produced no plan or credit drawdown.

OpenAI’s existing Codex plan documentation says consumption varies with the model, task complexity, context, reasoning, speed and tools. Consumer view brings those variables into a workload-level breakdown, giving users a way to investigate why one coding task consumed much more capacity than another.

The new system is designed to account for subagents

OpenAI’s current documentation warns that usage attributed to a Work or Codex chat may omit work delegated to subagents or run as separate tasks.

Build 7868 contains a newer accounting path that discovers subagent conversations associated with a root chat and attempts to request their usage as one group. The resulting measurements can then be attached to the original conversation in the dashboard.

The implementation does not guarantee complete attribution. Chats can be omitted when the application cannot fully identify their descendants, and delayed or unavailable usage records can leave gaps in the ranking. The interface warns users when some chats or usage are missing.

This distinction matters because OpenAI introduced the Codex desktop application as a place to run multiple agents in parallel. A single job may fan out across several agents and conversations. Grouping that work under its originating task gives the user a clearer picture of what consumed the allowance.

Usage history spans Work, Codex and other agentic tasks

Consumer view also contains seven-day and 30-day usage charts that can be grouped by product or model. The page covers Work, Codex and other agentic tasks while explicitly excluding ordinary ChatGPT conversations.

Tool activity tracks plugins called and skills used. A Code review tab can appear when recent activity contains reviews, comments or reactions. The interface warns that historical usage data is approximate and may be delayed by as much as six hours.

The history chart represents plan activity and eligible credit consumption. It is separate from the meter showing remaining capacity, and its presentation can vary with the data available to the account.

The ranking uses locally available chats

The feature starts with locally available chats that were active during the preceding 30 days. The application associates their identifiers with usage measurements returned by OpenAI and displays those totals beside conversation titles stored by the client.

RuntimeWire found no message bodies in the request assembled for this ranking. That finding applies only to this usage-accounting function and does not describe how OpenAI processes conversation content elsewhere.

Consumer view is scoped to the signed-in member and active workspace. RuntimeWire found no administrator-facing list of employee chat titles in the new route.

The Consumer view route and its per-chat rate-limit measurements were absent from the preceding Windows build reviewed by RuntimeWire. Build 7868 contains the interface, eligibility checks and supporting usage logic required for the feature.

OpenAI has already launched chat-level credit reporting for eligible enterprise workspaces. Consumer view tests a finer-grained version tied directly to the limits users encounter while working: which chat consumed the allowance, which settings contributed and whether associated subagents added to the total.

RuntimeWire asked OpenAI whether Consumer view is active for any customers, how widely it will be offered and whether the new subagent accounting is intended to close gaps acknowledged in the company’s current documentation. We will update this report if OpenAI responds.

Reporting record

RuntimeWire identified the changes by comparing consecutive production builds of Codex for Windows and reviewing the newly added interface and supporting application logic. No production usage request was sent.