Feature Request: Add a conversation-capacity meter and automatic handover warning to every ChatGPT chat

ChatGPT Feature requests
1

ChatGPT needs a visible conversation-capacity indicator for every chat.

The indicator should show:

  1. The approximate percentage of the current conversation context already used.
  2. The approximate remaining conversation capacity.
  3. Progressive warnings at approximately 70%, 85%, 95% and 100%.
  4. A clear distinction between:
    • conversation/context capacity;
    • account-wide message or model usage limits;
    • ChatGPT Work, agent, research and tool-processing limits.
  5. The currently selected model and reasoning level.
  6. Whether the chat is operating in normal Chat mode or Work/agent mode.
  7. A warning before a chat is moved into a different workflow that cannot later be returned to normal Chat.
  8. An automatic handover/export option before the chat reaches its limit.

The handover should preserve the project name, chat name, active instructions, decisions, workflow rules, files and sources, current status, last completed action and exact next action.

This feature is important because users currently have no dependable warning that a long-running project chat is approaching its limit. Users may only discover the problem when the chat suddenly becomes unavailable or loses continuity.

The meter may be an estimate because hidden instructions, tools, reasoning and attachments affect token usage, but an approximate warning would still be extremely valuable.

Please record this as product feedback for the ChatGPT product team.

2

Additional clarification to my existing feature request:

The ChatGPT Windows desktop app already displays account-level usage information, including:

  • remaining five-hour usage;

  • remaining weekly usage;

  • reset times; and

  • available usage resets.

The app’s settings also include a specific option called “Show context window usage.”

This suggests that ChatGPT already recognises or can expose context-window usage separately from general account usage. However, users still need a clear and reliable indicator inside each individual chat and project.

Please consider adding separate indicators for:

  1. Current conversation or project-chat context used and remaining.

  2. Account-level five-hour usage remaining.

  3. Weekly account usage remaining.

  4. Work, agent, computer-use, tool, and connector allowances.

The conversation indicator could be approximate because files, tools, hidden instructions, reasoning, and attachments affect usage. However, an approximate meter with warnings at 70%, 85%, 95%, and 100% would be far better than receiving no warning until the chat can no longer continue.

When a conversation approaches its practical context limit, ChatGPT should automatically offer to create a complete handover or export containing the project name, chat name, instructions, decisions, workflow rules, files, current status, last completed action, and exact next action.

The system should also warn users before moving a conversation into Work, agent, or another resource-limited mode if that change could affect whether the conversation can later return to normal Chat mode.

This is an additional clarification to my existing feature request, not a separate duplicate request.

3

I strongly support this, especially for ChatGPT Work.

One additional point from real-world use: conversation capacity does not seem to correlate simply with the number of visible messages.

In one recent Work session, a relatively fresh conversation handled a very large tool/text payload when the agent attempted to move a binary artifact as compressed base64 instead of using file-native transport. The resulting incomplete text artifact was about 29 MB. Shortly afterwards, that conversation reached its practical limit.

I cannot prove that the large tool payload was the sole cause, so I would not treat this as a confirmed implementation detail. But it illustrates why any capacity indicator should account for tool I/O, attachments and hidden execution context — not just visible chat length.

For Work, an early warning would be particularly valuable because the conversation may already contain completed execution steps, tool state, artifacts and expensive Work usage.

Ideally the warning could also trigger a structured handoff containing:

  • last validated state
  • completed steps
  • active files/artifacts
  • unresolved blockers
  • exact next step
  • things that should not be repeated

That would make approaching the conversation limit manageable instead of catastrophic.