The short answer
Package the ChatGPT session; do not narrate it. With Zambo added to ChatGPT, the session's calls run through https://zambo.dev/mcp and each step is saved with a verifiable receipt: 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. The handover tool closes the session into one link. In Claude, with the same endpoint added, the receiving session reads the package, checks the receipts it cares about, accepts, and continues the job from the recorded state. The reverse trip has its own page: handing work from Claude to ChatGPT.
How do I move a conversation from ChatGPT to Claude?
Move the work, not just the chat text. With Zambo added to both apps, package the ChatGPT session as a HANDOVER with its state, its decisions, and its receipts, then open that package in Claude, which reads the record and checks it before continuing. That keeps the history in one record both models can use, instead of a pasted summary that drifts at every hop. Add Zambo to your AI in both apps before the next move.
The manual transfer, honestly
The standard ChatGPT to Claude move is copy and paste at three possible depths, and each depth loses a different thing. Pasting the final outputs loses the path: Claude sees the drafts but not how they were reached, so it cannot tell a settled decision from a first guess. Pasting the whole conversation keeps the path and buries it: hundreds of turns of prose in which the actual tool results, the load-bearing facts, are scattered and unmarked. Asking ChatGPT for a handover summary is the best of the three and still a retelling, written by the session whose work it describes, dropping whatever that session judged unimportant, which is a judgment Claude never got to make. None of the three gives Claude anything it can verify. It starts from faith and spends its first exchanges rebuilding confidence you already paid for once.
The transfer, step by step, ChatGPT side first
- In ChatGPT: finish the current step, then package. The work so far ran through the endpoint, so its steps and receipts already exist outside the chat. Run the handover tool with the recipient named, the goal stated, and the next steps listed as you want Claude to receive them. You get one link.
- Inspect the package yourself. The link opens in a browser: goal, steps, saved results, receipt links, and anything marked as not checked. This is the exact object Claude will read. If a step is missing, it is missing from the record, and you learn that now rather than after the move.
- In Claude: add the endpoint, open the link. With Zambo added in Claude, the session reads the handover as data, not as a story: each step with its tool, its saved result, and its receipt pointing back to the source receipt.
- Claude verifies, accepts, continues. Receipts on the steps Claude will build on get recomputed at the public verify endpoint. Then acceptance lands on the record, and Claude's own steps chain onto the same trail, so the finished job reads as one job with a visible seam, not two half-jobs in two apps.
Setup for each side lives on this hub: adding Zambo to ChatGPT and adding Zambo to Claude. The general pattern, any AI to any AI, is switching AIs mid-task; the version where the receiver is an agent that signs for the package is agent-to-agent handover.
What does not transfer, said plainly
The handover carries work that was recorded. Chat in a ChatGPT conversation that never touched the endpoint produced no receipts, and the package cannot invent them after the fact; the honest scope line for the whole mechanism applies here too. What transfers is the recorded spine of the job: the tool steps, their results, and their proof. If part of the job lived only as conversation, include it the low-tech way, as a note in the handover summary, clearly a note, and let the recorded steps carry the weight they can prove. From the first recorded call onward, the split disappears: run the work through the endpoint and the whole job is transferable, every time, in both directions.
Scope, stated once: a verifiable receipt proves the saved result was not changed since it was saved. It does not prove the result was right. Claude inherits evidence about the ChatGPT half of the job, and judging that evidence is the first real work it does on arrival.
Questions, answered
How do I transfer a project from ChatGPT to Claude?
If the project ran through the Zambo endpoint in ChatGPT, package the session with the handover tool: the goal, the steps taken, the saved result and receipt for each step, and the next steps, at one link. In Claude, with Zambo added, open the link. Claude reads the steps as recorded, verifies the receipts behind the steps it will build on, accepts the handover, and continues from the recorded state. No transcript mining, no summary to trust, and the ChatGPT half of the job stays checkable afterwards.
What does Claude see when it opens the handover?
The job as a record, not as a chat: the goal, the sender's summary, the listed next steps, and each step with the tool that ran, the saved result, and a verifiable receipt whose published copy points back to its source. Steps the sender marked as not checked are marked as not checked. Claude can open any receipt at the public verify endpoint and recompute its check before relying on it, and the handover page itself reads in a browser if you want to inspect what changed hands.
Do I need Zambo added in both ChatGPT and Claude?
For the full path, yes, and it is the same endpoint in both: https://zambo.dev/mcp. In ChatGPT the recording happens because the session's calls run through it; in Claude the pickup happens because the receiving session can read the record, verify receipts, and accept the handover as tool calls. The handover link itself opens without any setup, so you can inspect a package in a browser before connecting anything.
Can Claude check what ChatGPT did before continuing?
Yes, and that check is the point. Each step's receipt commits to the exact saved bytes of that step's result. Claude, or you, recomputes the check at the public verify endpoint: a step that verifies is a step as recorded; a step that does not verify gets re-run instead of inherited. A receipt proves the saved result was not changed since it was saved, not that the result was right, so Claude still judges the work, against the record of the work that actually ran.
What does a ChatGPT to Claude handover cost?
The handover tool is free. The work on both sides runs on the same terms: 20 free calls per day, then a $0.99 Day Pass for 24 hours of higher limits, or the Zambo Pass at $29 for 30 days, one payment, $290 for 365 days, and $19 for 30 days for $ZAMBO holders. One endpoint covers both apps, so the pass you already have is the pass that carries the job across. People pay by card through PayPal; agents pay in USDC on Base through x402.