Direct answer

You closed the laptop, the app crashed, or the machine restarted. Your AI session is gone. Here is how to resume without starting over.

The panic is specific: hours of work were in that session, the window is closed, and reopening the app shows a blank chat or nothing at all. Take a breath and separate two things that feel like one loss. The conversation may be gone. The work, meaning the calls that ran and the results they returned, may not be, if anything outside that window was keeping the record. This page is the recovery order: what to check, in what sequence, and how to restart so it cannot happen the same way twice.

Start the recovery from the record: any Zambo receipt can be opened and checked free at zambo.dev/verify, with no signup. If the lost job ran through Zambo, its receipts are still there even though the session is not.

Updated October 24, 2026.

What is actually lost, and what usually is not

The conversation thread. In many apps this is the fragile part. A crash, a closed laptop with a dead battery, a browser reset, or a timed out session can take the visible thread with it, and with it the agent working context.
Files the work produced. Anything written to disk, a document, a shared folder, or a repo before the crash usually survives. Check the actual save locations before mourning them.
Calls that ran through a recorded endpoint. These survive by design. On Zambo, every call saved its result when it ran and returned a receipt, stored outside your laptop session. The session dying does not touch them.
Unfinished, unsaved chat output. Text the agent produced in the chat but that was never saved anywhere else is the real loss in most crashes. Be honest about which bucket each piece of work is in; the recovery differs.

The recovery order

1. Do not start a new attempt yet. A fresh agent redoing the job from memory creates a second, conflicting version of everything. First find out what survived.

2. Check the save locations. Documents, downloads, project folders, repos, sent email, and any shared drive the work touched. Open the newest versions and note their timestamps. That timestamps list is your first map of how far the job got.

3. Read the call record back. If the job ran through Zambo, pull the receipts for the session in order. Each receipt holds the saved result of its call, so the sequence tells you exactly which steps completed, what each one returned, and where the job stopped, including the last successful call before the crash. A call with no receipt after it is where the thread ends.

4. Write the resume brief from what survived. Goal, steps completed with their outputs named, decisions already made, the exact stopping point, and the next step. Build it from files and receipts, never from your memory of the chat. Your memory of a lost session is as partial as the agent would be.

5. Resume in a fresh session, loaded with the brief. Hand the brief to the new session, or close the recovered state as a handover and let the next agent continue from the recorded steps. Check the first new outputs against the brief before letting the run build speed. The method is in AI agent context handoff.

6. Close the recovered job properly. Once the work is finished, close it with a handover so the final state, not the crash, is what the record shows.

Why the record survives when the session does not

A chat session lives in one app, on one machine, in one window. Everything about it is local and temporary. A per-call record lives outside it. Zambo is the verifiable work layer for AI: when a call runs through the Zambo MCP endpoint, the result is saved as the call runs and a receipt is issued for it, on Zambo side, not inside your chat app. Close the laptop, lose the battery, restart the machine, switch to a different AI entirely, and the receipts are still readable at their links, checkable by anyone at zambo.dev/verify.

That separation is the whole recovery story. The session is a workspace. The record is the filing cabinet. Losing the workspace is annoying. It is only a disaster when the filing cabinet was inside it. The related failure, where the session survives but its memory fades mid-task, is covered in your agent forgot mid-task.

Set up so the next crash costs minutes, not the job

Run important work through recorded calls from the start. Recovery is only possible for calls that left a record. Retrofitting a record onto a lost chat is not possible.
Save outputs outside the chat as you go. Every finished stage lands in a file, a doc, or a handover at the moment it is finished, not at the end of the day.
Close stages with handovers on long jobs. A handover every stage means the worst case after any crash is redoing one stage.
Before resuming a long unattended run, set the guardrails. If the job will run while you are away from the machine, pair the recovery setup with an alert before the agent overspends and the caps in stop an AI agent overspending, so a resumed run cannot quietly burn budget while you are not watching.

Honest limits

A record recovers what ran through it. Chat text that never became a call, a file, or a saved output is gone with the session, and no tool brings it back. Part of this page is a triage: recover the recorded work fully, name the unrecorded part plainly, and redo only that part.

A receipt also proves a narrow thing: the saved result was not changed after it was saved. It does not prove the result was right, and recovery still needs your judgment on the outputs you are resuming from. Read them as you rebuild the brief. Resuming confidently from a bad early output just reaches the wrong answer faster.

Set up the record before you need it

Crashes are not rare events to fear; they are normal events to design for. Run the job through Zambo from the first call, save each stage outside the chat, and a dead laptop becomes a ten minute interruption: read the receipts, write the brief, resume.

Work run through the Zambo MCP endpoint at zambo.dev/mcp (153 tools, 35 on the default surface) saves its result when the call runs and returns a verifiable AER-1 receipt (draft-zambo-aer1-26) for every call. In short, an agent receipt is a record of one ai agent tool call that anyone can check: it keeps the exact saved bytes and their sha-256 digest, and it proves the result was not changed since it was saved, not that it was right. Start free: 20 free calls per day, then they upgrade. The Day Pass is $0.99 for 24 hours. Checking any receipt is always free at zambo.dev/verify.

Questions, answered

My laptop died mid-task. Is all the AI work gone?

The chat thread may be, but the work often is not. Files saved to disk or shared storage survive. More importantly, any call that ran through a recorded endpoint like Zambo left a receipt with its saved result, stored outside the session. Recover in order: check save locations, read the receipts back in sequence, write a resume brief from them, then restart from the brief.

How do I find where the job stopped before the crash?

Read the call record in timestamp order. The last receipt in the sequence is the last completed call, and its saved result is the true stopping point. Anything the agent planned to do after that call never ran, whatever the plan said. On Zambo each receipt can be opened from its link and checked free.

Can I resume the same work in a different AI app?

Yes, and this is often the cleanest recovery. The record does not live in the crashed app. Write the resume brief from the files and receipts, hand it to a session in any app connected to the same endpoint, and the new agent continues from the recorded steps rather than from a retelling.

What if nothing was recorded and the session is just gone?

Then recover the old way: files on disk, sent messages, downloads, and your own notes, and redo the rest. Be honest about that cost, then change the setup before the next long job: recorded calls, outputs saved outside the chat, and a handover at every stage, so the next crash has a record to recover from.

Does a receipt prove the recovered work is correct?

No. It proves the saved result is the result that was returned at the time, unchanged since it was saved. Correctness is still your call when you read the outputs during recovery. What the record removes is doubt about which version survived and where the job stopped.

Start where the work starts: run the next job through zambo.dev/mcp, close it with a handover, and check any receipt free at zambo.dev/verify. If you only need a day, the Day Pass is at zambo.dev/day-pass.