Direct answer

ChatGPT hit its usage limit in the middle of my work, or the session just died. What do I do?

ChatGPT work dies three ways: the usage limit lands mid-task, the stream interrupts and never resumes, or the session crashes about an hour in and the reload comes back with the context gone. All three are the same accident: the state of the job lived only inside the session. This page is the rescue in order: salvage the state while the thread is still reachable, carry it to another AI or a fresh session as a handover record, then set up so no dead session can take the work again.

The end state of the fix, live: hand a job over with its record intact, and check any record free at the public verifier, no signup.

Updated October 20, 2026.

Step 1: Salvage the state while the thread is still reachable

Reload once. Check your history. If the thread opens, do not continue the work; spend your first message on the state. Paste this:

"Stop working on the task. Write a complete state summary of this job for someone who has never seen this conversation: the goal, every decision you made and why, every output you produced and its current state, what is finished, what is half-finished and exactly where it stopped, and what remains to be done, in order. Include anything you were about to do next. Be specific enough that another AI could continue from this summary alone."

Save two things outside the chat, on your own machine: that summary, and every output file or text the session produced. For Codex and coding work, the files are the state that matters most: get every produced or changed file saved before anything else, because the conversation is the least durable part of the job and the files are the work itself.

If the thread will not reopen, or it reopens hollow, work from what survives: the history view, your downloads, your saved files. A fresh session can rebuild an approximate state from the transcript, but be honest with yourself about what that is: a summary of what was said, not the working context the dead session had. This failure mode, a job dying about an hour in with its context gone on reload, is reported by ChatGPT users constantly, and it is not bad luck. It is what happens when the transcript is asked to be the storage.

Step 2: Carry the job across with a handover record

A state summary pasted into a new chat works, but it is the weakest way to move a job: text gets trimmed, files get forgotten, and a day later nobody can tell what the receiving AI was actually given. The stronger move is a handover: the goal, steps, outputs, and open items travel as one record. The receiving session, whether that is a fresh ChatGPT session, Claude, Gemini, or Grok, starts from the record, checks it, and accepts or flags what it finds, so gaps surface at the handover instead of three hours later.

That is what Zambo handover is built for: one link carries the whole job, and every step inside it carries its own record, produced when the work ran, that anyone holding the link can check free at zambo.dev/verify. The full pattern is walked through on the hand work over with proof page, and the general move is covered on the continue in another AI page.

Step 3: Set up so a dead session never costs the work again

Limits, stream failures, and crashes differ in cause and are identical in effect, and no setting on your plan removes any of them. The variable you control is where the job lives. A job whose state lives inside one chat session has a single point of failure; a job whose state is saved outside the session, as the work happens, survives every one of these deaths.

When a job runs through Zambo, every tool call is saved as a record at the moment it runs, and the job can be handed over mid-stream at any point: before the limit, after the crash, or a week later, to any AI, with the open items intact. The session becomes replaceable, which is the correct relationship to have with it.

For $0.99, a Day Pass covers 24 hours of recorded work: get the Day Pass and run your next real job with the net on. Ongoing use is the Zambo Pass at $29 a month or $290 a year, and $19 a month for $ZAMBO holders, which adds private handover links, your branding, and the approval record as PDF and JSON files.

Honest limits

Three edges, stated plainly. First, some sessions cannot be reopened at all, and when that happens before any salvage, recovery is limited to history, downloads, and whatever you saved along the way: prevention is the only reliable answer to a session that dies unwitnessed. Second, nothing here raises your ChatGPT limit, changes your plan, or makes streams more stable: those are your plan's terms and the service's health, and the only lever you hold is what a failure costs you. Third, a handover record preserves the state of the work and proves the saved results were not changed after the fact; it does not prove the work in them was correct. Check the outputs when the job resumes. Why records settle questions that transcripts cannot is laid out on what counts as proof that AI work happened.

Frequently asked questions

ChatGPT hit its usage limit in the middle of my work. What do I do now?

Do three things, in this order. First, salvage the state: while the thread is still reachable, make ChatGPT write out the state of the job, the goal, the decisions made and why, every output produced so far, the current state of each, and what remains to be done in order, and save that summary and every output file outside the chat. Second, carry the state to a fresh session or another AI as a handover record, so the next session starts from the record instead of from your memory of the conversation. Third, before the next big job, set the work up so its state lives outside any one session. The limit, the crash, and the dead stream all cost you the work the same way: the state of the job existed nowhere but inside the session that just closed.

ChatGPT said streaming interrupted and my session died. Is the work gone?

Often not all of it, but do not count on the thread alone. Reload once and check your history: the conversation frequently survives as text, and anything you already copied out or downloaded is safe. What usually does not survive is the working context: reload the thread and the AI may have the words of the conversation but not the grip on the job it had an hour in, and some sessions cannot be reopened at all. Use the surviving thread immediately to rebuild a state summary, save it outside the chat, and treat the thread itself as a transcript of what was said, not proof of what was done.

How do I move a half-finished job from ChatGPT to another AI?

Get the state out of ChatGPT first: one complete summary covering the goal, decisions and reasons, outputs so far with their current state, and the remaining work in order. Then hand that state to the receiving AI, whether that is Claude, Gemini, Grok, or a fresh ChatGPT session, as the opening brief, with the output files attached. A plain paste works in a pinch, but it carries no record of what was already done. A handover record carries the steps, outputs, and open items as one checkable package the receiver accepts or flags, so nothing silently drops between the two AIs.

Does reloading the chat bring the context back?

Sometimes the thread comes back; the depth of context often does not. A long session accumulates working state that is never fully written into the visible messages: the plan, the rejected options, the reason the current draft looks the way it does. On reload, the AI rereads a transcript and reconstructs an approximation, which is why a resumed session can feel like a new hire reading the minutes. Reloading is worth one try. After that, work from a state summary you control, because the transcript was never the job. The job is the state, and the state needs its own home.

How do I stop a dead session from costing me the work next time?

Run the job so its state is saved as it goes, not reconstructed afterwards. When work runs through Zambo, every tool call is saved as a record at the moment it runs, and the job can be handed over mid-stream, to another AI or a fresh session, with steps, outputs, and open items intact, whether the session ends by limit, crash, or closed laptop. For $0.99, a Day Pass covers 24 hours of recorded work, which is enough to run your next ChatGPT or Codex job with the safety net on. Ongoing use is the Zambo Pass at $29 a month or $290 a year, and $19 a month for $ZAMBO holders. Anyone you hand the record to can check it free at https://zambo.dev/verify/ with no signup.