The definition
The short answer
An AI session handover is a working session packaged so it can be continued in a different app: the goal, the state of the work, what is done with a verifiable receipt for each finished step, the exact next step, and every decision made along the way, carried in one checkable record. Another AI, or a human, opens that record and continues the job from it. What moves is the job itself, not a copy of the conversation about the job.
The same practice is often called a session handoff; both terms name the same packaged, receipt-carrying record.
The phrase also gets used for tools that move a chat from one app to another. Moving the chat is the easy half. The definition above is the one that matters when the work has to survive the move: if the receiver cannot tell what state the job is in, what was already decided, and what already ran, the session did not really transfer. A scrollback did.
What a session actually holds
- The goal, and the current state. What the job is, and precisely where it stands: finished, half-finished, or blocked, and blocked on what.
- Every decision, with its reason. The choices the session already made and settled. Written down, they stay decided; left in the scrollback, the next AI re-argues them.
- The partial outputs. Drafts, files, queries, and intermediate results: the material the next step builds on, not only the polished final answers.
- What is done, with receipts. A verifiable receipt for each finished step, saved when the step ran, so the done part can be checked in the new app exactly as it was in the old one.
- Sources checked, and sources not checked. The honest half of the record: what the work rests on, and what nobody verified, named as not checked instead of smoothed over.
- The exact next step. Stated before the session ends, so the receiver starts by doing, not by reconstructing.
Why a copied transcript fails the receiver
A chat export looks like a complete transfer and behaves like a bad one:
- It buries the state. The decision that matters sits deep in pages of back-and-forth, surrounded by dead ends that were tried and dropped. The receiving AI reads all of it with equal weight, or skims and misses the one line that changed the plan.
- It is narration, not a record. A transcript shows what was said about the work. It cannot show what the tools actually did, and nothing in it can be checked after the fact.
- It reopens settled questions. With no decision list, the next AI happily re-argues choices the session already made, and quietly reverses some of them.
- It drops the unfinished edges. Partial outputs, the step that failed, the source nobody checked: a tidy recap smooths these over, and the receiver inherits the gaps without knowing they exist.
The handover is the transfer unit
In Zambo, the package a session travels in is a handover: the goal, what is done, what is next, and every decision, with the receipts attached, closed by a summary receipt. The session handover is the act; the handover is the artifact that makes the act checkable. Open the handover's link in Claude, ChatGPT, Gemini, Grok, or whichever app is next, and the receiving AI continues from the record: it reads the state, honors the decisions, checks the receipts it wants to check, and takes the stated next step.
That is also the answer when the move is forced. A usage limit ends a session on the app's schedule, not on yours. If the session's record lives outside the app, the limit costs you a switch, not the job: the same link opens in the AI you move to, and the work continues from exactly where it stopped.
A record built to be received: a live example
An outside agent studio wired AER-1 receipts into its standing-check deliveries, so the record of its sessions' work reaches the next pair of hands with the evidence attached instead of promised. One of its delivery records is public: receipt 28c79cde, whose saved bytes are 2,078 bytes long and recompute to exactly the output hash the record commits to, sha256:5f06acfb. Anyone can rerun that check; it needs no account and no trust in the studio or in Zambo.
The same record also shows the part a transcript would hide: its own audit flags three of its claims as weakly evidenced, and reports issuer authentication as not checked. A receiver starting from this record knows what is proven, what is claimed, and what is missing, before doing any new work. That is what a session handover has to carry.
How to hand off a session
- Run the work where it leaves a record. Work run through Zambo's 153 tools saves a verifiable receipt per step as it runs, under AER-1, an open IETF Internet-Draft (draft-zambo-aer1-23). The record creates itself; nobody writes it up afterward.
- Set the trigger inside the session. Save this line where your AI keeps its instructions: "When I say WRAP UP, hand this job over with Zambo: the goal, what is done, what is next, every decision, with the receipts attached, and give me the continue link." From then on, saying wrap up is the whole handover procedure: the AI packages the session and hands you the link.
- Open the link in the next app. Paste it into Claude, ChatGPT, Gemini, or Grok, or hand it to the agent you are moving to, and it continues from the record. When the switch is forced by a usage limit, start at zambo.dev/continue; the same record carries the work across.
Checking a handover is free, for you and for whoever receives it. The recording runs on the free tier on a normal day; a $0.99 Day Pass covers a heavy day of the full machine, and the $29 Pass ($19 for $ZAMBO holders, $290 for the year) fits work that moves every day.
Questions, answered
How do I hand off an AI session to another app?
End the session by packaging it, not by copying it. In Zambo, the trigger is one saved line: when I say wrap up, hand this job over with Zambo: the goal, what is done, what is next, every decision, with the receipts attached, and give me the continue link. Say wrap up, take the link it returns, and open that link in the other app: Claude, ChatGPT, Gemini, Grok, or another agent. The receiving AI reads the record, checks whichever receipts it wants to check, and continues from the stated next step.
What is a session handover?
A session handover is a working session packaged so it can be continued in a different app: the goal, the state of the work, what is done with a verifiable receipt for each finished step, the exact next step, and every decision made along the way, carried in one checkable record. Another AI, or a human, opens that record and continues the job from it. What moves is the job itself, not a copy of the conversation about the job.
What does an AI session actually hold?
More than its messages. A working session holds the goal and the current state of the job, every decision made along the way with its reason, the partial outputs the next step builds on, the sources that were checked and the ones that were not, and, when the work runs through a system that records it, a verifiable receipt for each finished step. A transfer that moves only the messages leaves most of that behind.
Why not just copy the chat transcript into the new app?
Because a transcript is narration, not a record. It buries the decisions inside pages of back-and-forth, gives dropped dead ends the same weight as the final plan, shows what was said about the work rather than what the tools actually did, and contains nothing the receiver can check. The new AI has to reconstruct the state, re-argue settled decisions, and inherit gaps it cannot see. A session handover moves the state, the decisions, and the receipts, so the receiver starts from the record.
What does the receiving AI get in a session handover?
One link that opens the whole record: the goal, the state of the work, what is done with a verifiable receipt for each finished step, the exact next step, every decision with its reason, and the items named as not checked. The receiver does not have to trust the sending session's recap, because it can recompute the receipts itself, and anything that cannot be verified is labelled as unverified instead of hidden.