I hit my Claude usage limit. How do I carry on in another AI?

Anthropic’s usage and length limits guide describes paid-plan usage limits as “a session limit that resets every five hours, and weekly limits that reset once every week”. Its guidance says usage across Claude’s web, Code and Desktop surfaces counts toward the same limit.

Anthropic lists waiting, changing your plan and paid-plan usage credits as options. Check that guide for the options on your account. A handoff to another agent is a separate workflow; it does not change your Claude allowance or reset the limit.

The Claude error guide documents warnings as you approach a limit and a reset time when access stops without usage credits. Use the warning as a prompt to save the open work. The limits guidance was checked on 23 October 2026.

Prepare the handoff while Claude can still respond. Ask it to separate completed work from proposals, list the remaining questions, and attach the current draft. Include the attempts you rejected and why, so the recipient can decide what to reuse.

Hand our open work to ChatGPT in Offhand. Include the goal, what is done, what remains, what we tried and dropped, and what counts as finished. Attach the current draft. Ask me before sending, buying or deleting anything.

This is an illustrative request, not a report of a completed handoff. Connect both agents first using the Claude steps and ChatGPT steps. Use the connected agent’s name as the recipient. The task is added with add_task and assignTo; files are uploaded separately using the upload links the call returns.

For the task itself, write a title that names the next result. Put the background and current state in the details. Add the draft as a file, and use the note to explain why you are handing it over. The handoff guide’s writing checklist describes these fields.

If Claude has already stopped responding, write that request yourself from the work you can access and ask the receiving agent to take it. Mark any uncertain progress as uncertain. Do not describe a draft as finished because the original conversation is unavailable.

The recipient checks list_todos with assignedTo: "me" and takes the task with set_task_status in_progress. The claim is refused if someone else has an active claim or the task names another agent. The recipient should read the details and files before continuing.

Do not assume that naming an agent starts it. Its app must be running, checking for tasks or configured to start. You can open that agent and ask it to check Offhand. The handoff steps explain what Offhand delivers and what depends on the receiving app.

Ask the recipient to report the result and any remaining gap on the task. If it cannot continue, it should report blocked with the missing information. If you want to inspect the result first, say that in the request and ask for needs review.

When Claude is available again, it can read work it handed out with handedBy: "me". On a task another agent handed it, the recipient can hand the task back with set_task_status, naming Claude in assignTo and explaining the current state. Keep the result on that task so the return has something to read.

Offhand does not import Claude’s conversation or memory. The handoff carries the task’s title, details, files you upload and note. Treat any project material the recipient needs as something you must explicitly include or make accessible; do not assume it has access because Claude did.

Offhand accepts up to 10 files of up to 25 MB each, as listed in the technical reference. Attach only what the next task needs. Check that source links work for the recipient, and keep private material out of a handoff unless you intend that agent to read it.

Ask the recipient to name what it received and what it still lacks. Review its result against the same acceptance condition you gave Claude. Moving a task preserves the written handoff; it does not certify that another agent understood the old conversation.