# Manage conversations & history

Keep task context focused, find history, adjust project membership and restore archived conversations.

Updated: 2026-11-02
Desktop 0.5.0-rc.60 · Web 0.5.0-rc.36
Canonical: https://ai.z-group.cn/docs/en/conversations/

## Give each conversation a clear outcome
Create a conversation within its project and use an object-plus-task name, such as “Entrance stone slip-test review”. Start a separate conversation when the sources or objective change substantially.
## Everyday actions
1. Rename, set status, add labels or flag the conversation using its menu.
2. When changing project membership, choose the destination and verify its position in the list.
3. Use project/conversation search to locate history. Long histories may load in batches.
4. Archive inactive work. Find and restore it in Settings → Archived conversations.
## Share a conversation to the browser
Use Share in the conversation menu to create a read-only web link (ai.matlook.com/s/…). Anyone with the link can follow the full conversation in a browser without installing or signing in. Links stay valid for 30 days. Refresh share updates the content and restarts the clock; Stop share invalidates the link immediately.

Sharing uploads a text snapshot of the conversation and does not include local attachments. Conversations that were never shared stay on your computer and are not uploaded. Review the conversation for client data or account credentials before sharing; the link is readable by anyone who obtains it.

## If a conversation is missing
Check the execution target and workspace first, then project, status and label filters and the archive. A project's list does not contain all conversations. All conversations also includes tasks with no project.
## Continue with useful context
State the previously agreed outcome, new sources and requested changes. For long histories, ask for a summary of evidence and unresolved items, then review it yourself. Context compaction organizes the model's context; it is not a backup of your original files.
## Is the task complete?
Generation ending is a runtime state. You set the business status after checking the output. Done does not mean the result was sent to a client or approved for procurement. Stopping does not undo completed actions.

## Submit a plan and continue after approval
Starting with rc.77, plan submission waits until the review card is ready before ending the turn. Check its scope before approving continuation. Model generation ending does not constitute your approval.

A failed plan read or handoff produces a tool error and retains the saved plan. Upgrading does not replay historical tools; a stale turn cannot acknowledge a newer submission. If an older conversation shows an unknown-ending error without a review card, ask the AI to resubmit the existing plan for review only, then approve after checking it. Do not directly repeat operations such as cleanup.