Sync and share team memory

onemind keeps a shared memory for your team. Any member can propose additions, and admins approve them. Approved proposals become part of the shared memory that every team member's AI tools receive.

Sync and verify

Running onemind sync syncs your local memory with the server and retries any queued proposals that were not sent on the previous attempt. After sync completes, the CLI reports the current sync status.

Propose team memory

To propose an addition to team memory from the CLI:

onemind propose "Your proposed memory text here"

The proposal is sent to team admins as a notification. An admin can approve or reject it from the Notification inbox in the onemind BETA app, or from the shell with onemind approve <id> or onemind reject <id>.

When an admin approves a proposal, onemind publishes a new version of shared team memory. All team members receive the update on their next sync. Rejecting a proposal leaves shared memory unchanged.

For longer Markdown, use onemind propose --file PATH. Proposals stay pending until an administrator approves them.

Explicit proposals do not require automatic analysis. Knowledge analysis is off by default; while it is off, onemind does not scan skills or consume queued events as analysis input. Explicit onemind propose items still use their separate reliable-delivery queue. Independently of analysis, the generic reconciler may upload valid structured onemind proposal events, including events from Cursor, Qoder, TRAE, and WorkBuddy roots, but only for the exact user-activated Project binding whose upload permission is on.

Automatic analysis starts only after the user opts in. onemind then reads only changed Codex SKILL.md files from documented skill roots and structured onemind work-history/queue items selected by active bindings. Apart from read-only WorkBuddy workspace/session-path metadata used for candidate discovery, onemind does not use host-owned database records, read proprietary chat/message content, or infer durable knowledge from arbitrary file changes. Analysis never expands Project-binding authority or turns a host-owned data store into analysis input.

The default kind is prompt_observation. A custom kind must contain only lowercase letters, numbers, hyphens, or underscores. Proposal input is limited to 1 MiB.

On success, onemind prints the proposal ID and explains that admin review is pending. If the server is temporarily unavailable, the proposal remains safely queued and the CLI tells you to run onemind sync to retry. Approval—not upload—publishes the entry into shared memory.

For the admin review workflow, see Team administration.

Auto-generate Minds from unread work history

The Mind Generator reads unread work history and creates a Mind pull request from it. When a new Mind draft is ready, a Mind draft generated notification appears in the Notification inbox.

Open the notification to review the draft. It goes through Safe Check and review like any other Mind update, and an admin publishes it with Approve and publish (see Minds and review).

For details on how to use the onemind agent for this and other memory operations, see The onemind agent.