BETA plugin workflows
The shared onemind plugin is BETA-first across all supported hosts. In addition to status, sync, and approved memory, its skill covers Projects, Matrix files and replacement approvals, Project skills, proposals, analysis, server-side Mind Check, instruction-file backups and Mind-sync restore points, integration lifecycle, service controls, and signed-update guidance. Alpha remains legacy; a plugin or CLI update does not replace the Alpha app.
OpenClaw, Pi, and Hermes expose onemind_read and onemind_action, alongside the existing
onemind_status, onemind_sync, and onemind_memory. Select an operation, pass separate
arguments strings, and set the exact absolute workingDirectory. Read/action separation does
not grant permission: actions must match the user's explicit request. Other hosts use the shared
skill with their supported MCP/terminal tools. Prepared/manual and Matrix-only limits still apply.
files.check may request approval, and mind-check uploads selected evidence; neither is a
passive read. Restore only an inspected, explicitly selected backup. Failed commands retain
stderr and failure status even after printing progress; timeouts and partial output never prove
success. Management actions without a published CLI/MCP operation remain in the BETA web/app.
onemind agent
The onemind agent is a cloud-only team agent available in Matrix on all active accounts. Start a conversation from any Matrix task to create tasks, assign work to AI tools, answer pending tool interactions, manage Minds, and request sync. The agent never runs code, edits files, or executes commands on a device; all work happens in the assigned tool.
The agent exposes 28 catalog operations covering tasks, Projects, memory, Minds, sync, and analysis. The interactions.answer operation lets the device owner respond to a pending tool question or permission from a running task without stopping it; a question also accepts the task creator. Operations that name a record are proposal-bound: the agent resolves the exact record before presenting the action card for review. Use /answer in the Matrix composer to reach interactions.answer directly. For a complete reference, see The onemind agent.
Create and update BETA Minds
Use onemind minds list --json and onemind minds show MIND_ID --json to inspect first-class Minds.
Create with onemind minds create --id ID --name NAME --key KEY --purpose TEXT --file PATH --idempotency-key KEY [--project PROJECT_ID] --json. Admin creation makes content active immediately; omitting the Project leaves it unassigned.
Update with onemind minds update MIND_ID --file PATH --title TEXT --expected-version N --expected-source-hash HASH --json, preserving mind.headVersion.version and mind.headVersion.sourceHash from the read. Optional --details and --file-name describe the proposal and candidate filename. A Project-bound Mind is required; updates submit Safe Check/review and do not publish. Reuse creation keys and exact update payloads on ambiguous retries; stop and resolve stale-head conflicts. Content files must be UTF-8; creation README is limited to 8,000 characters and updates to 512 KiB.
The plugin exposes minds.list, minds.show, minds.create, and minds.update. MCP equivalents are one_mind_list_minds, one_mind_get_mind, one_mind_create_mind, and one_mind_update_mind, with separate onemind.minds.read and onemind.minds.write consent. Matching server/CLI/plugin publication and reconnection are required before existing installations can use the new capabilities.
BETA 0.4.15 includes these four Mind tools in the fifteen-tool MCP source inventory. Existing connections need fresh onemind.minds.read / onemind.minds.write consent. Reconnect if the host has not refreshed its tool list or reports missing scope; publication alone does not prove the state of your active session.
MCP tools and permissions
The onemind MCP server exposes tools that let connected AI tools read status and memory, request sync, manage projects, and check project file assignments. Each tool requires one consent permission; grant only the permissions your workflow needs. For authorization steps, see Sign-in and accounts > Authorize the CLI or an AI tool.
| Tool | What it does | Permission required |
|---|---|---|
| Get sync status | Reads the current device sync and service state | Read machine and AI-tool sync status |
| Get memory | Reads the approved shared team memory | Read your Matrix's approved AI memory |
| Request sync | Asks connected devices to sync | Request sync on your connected devices |
| List projects / Get project | Lists or reads project details | Read Matrix Projects and pending edits |
| Create project / Update project | Creates a project or submits a project update proposal | Create Projects and submit Project edits |
| Decide project update | Approves or rejects a pending project update proposal | Approve or reject pending Project edits |
| List Minds / Get Mind | Lists or reads Mind details and versions | Read BETA Minds, approved content, and pending updates |
| Create Mind / Update Mind | Creates a Mind or submits a Mind update proposal | Create active BETA Minds immediately and submit content updates for review |
| Check project file freshness | Compares a local file's hash against the remote assignment | Check assigned Project file freshness and refresh status |
| Get project file refresh status | Reads the governed status of a pending refresh request | Check assigned Project file freshness and refresh status |
| Request project file refresh | Creates a device-scoped approval request to replace a stale file | Request approval to refresh an assigned Project file |
None of these MCP tools write local files directly. File replacement requires a separate approved workflow on the destination device.
Plugin operations for OpenClaw, Pi, and Hermes Agent
OpenClaw, Pi, and Hermes Agent dispatch all OneMind operations through two plugin tools: onemind_read for read operations and onemind_action for action operations. Select an operation by name, pass separate arguments strings, and set the exact absolute workingDirectory. Read/action separation does not grant permission: actions must match the user's explicit request.
The plugin provides 23 read operations and 34 action operations, for 57 operations total.
Read operations (23) cover: help and version information, device status, session status, team memory, review queue, update check, update instructions, service status, sync analysis status, sync status for the device and team, project list and detail, file list, skill status and resolution, Mind backup list, sync-backup list, integration list and status, and Mind list and detail.
Action operations (34) cover: sync, analysis control (on, off, now), service management (install, start, stop, uninstall), project create, edit, approve, and reject, file check, upload, download, and delete, skill install, update, and uninstall, Mind restore from backup, sync-backup restore, upload evidence for a server-side Mind Check, event proposal, proposal approve and reject, detect a selected ordinary Project target (it remains inactive until selected in OneMind), session verification, integration install, update, repair, and uninstall, and Mind create and update.
files.check may create or reuse a server-side approval request, and mind-check uploads selected evidence; neither is a passive read. Restore only an inspected, explicitly selected backup. Management actions without a published CLI, MCP, or plugin operation remain in the BETA web or app.