The onemind agent
The onemind agent is a cloud assistant built into Matrix that lets you create and assign tasks, read Projects and team guidance, work with Minds, check sync, and run the Mind Generator through a conversation. The onemind agent itself runs in the cloud; sync requests and Mind Generator runs are carried out by your own devices. Team admins can also create Projects, invite people, and run Safe Check on a Mind.
Start a conversation
Open the Matrix page and select + at the top of the panel. A new chat opens on the right with the heading What would you like to do?
When you open a new onemind agent conversation, four suggestion cards appear to help you get started:
- Create a task
- Assign or hand off work
- Connect an AI tool
- Check and sync context
Select a card to pre-fill the input with that request. You can also type freely or use Create manually to open the task creation form directly without going through the onemind agent.
Project picker
A Choose a project selector appears in the composer when no project has been selected for the conversation. Select a project to give the onemind agent context for the tasks and tools it shows.
Reasoning effort
When the model connection supports it, an effort selector appears in the toolbar. Select an effort level to control how thoroughly the onemind agent reasons through your request. The available levels are: None, Minimal, Low, Medium, High, Extra high, and Maximum.
The onemind agent identifies itself as onemind: "Replies come from onemind. Use @ or Handoff to involve another AI." In the composer, type @ to mention a team member or AI agent, and / to insert a Mind or one of the operations listed below.
Reply actions
Each reply shows a Copy button and a thumbs-up / thumbs-down rating. Use the rating to give feedback on a response.
The composer toolbar shows how full the conversation's context window is. Open Usage details to see What fills it:
| Category | What it includes |
|---|---|
| Conversation | Your messages and onemind's replies |
| Actions and results | Operations onemind ran and their results |
| Team guidance | Approved team memory |
| Projects | Project context |
| onemind commands | The operations onemind can use |
| Task and Team | The task and team details |
At 100%, onemind can't reply; start a new chat to continue.
What the onemind agent can do
The onemind agent can perform the following operations. Some, such as proposing a Mind update, create a pending approval that an administrator reviews before it takes effect.
Tasks
| Operation | What it does |
|---|---|
| List tasks and ready agents | Show current tasks and available AI tools in the selected Project |
| Define or refine this task | Create or update a task description and run its Task Checker |
| Assign or hand off this task | Set or change the AI tool assigned to a task |
| Run the Task Checker | Check whether a task has duplicate or similar existing work |
| Read this task's run history | Show the execution history for a task |
| Send a follow-up to a finished run | Continue a completed native run that can accept a follow-up |
| Archive or unarchive a task | Move a task to or from your archive |
| Edit a task | Revise the title or details of an existing task; saving creates a new revision and withdraws the current check or assignment |
| Resolve a Task Checker result | After the Task Checker flags potential duplicates, decide to create or discard the candidate task |
Projects and memory
| Operation | Who can | What it does |
|---|---|---|
| List Projects | Any member | Show the team's Projects |
| Read a Project detail | Any member | Show full details for a specific Project |
| Read approved Team guidance | Any member | Retrieve the current shared team memory |
| List Minds | Any member | Show available Minds |
| Read a Mind | Any member | Show an approved Mind and its review history |
| Create a Mind | Administrators | Create a new active Mind immediately; there is no review step |
| Propose a Mind update | Any member (Admin approval required) | Submit a change to an existing Mind |
| Create a Project | Administrators | Create a Project right away from its name, goal and description; nothing is assigned |
| Invite a person to the Team | Administrators | Create a 7-day invitation code and link to copy and share; onemind sends no email |
| Run Safe Check on a Mind | Administrators | Check an approved Mind for secrets, unsafe instructions and conflicts with other guidance; nothing is changed |
Sync and analysis
| Operation | What it does |
|---|---|
| Check device and tool sync | Show the current sync status for your devices and AI tools |
| Sync your devices now | Trigger an immediate sync |
| Show AI tool connections | List the AI tools connected in this Project and show how to connect another one on your Mac |
| Check the Mind Generator | Show the status of the Mind Generator |
| List Mind Generator runs | Show your recent Mind Generator runs |
| Run the Mind Generator now | Start a Mind Generation analysis (see Generate Minds with the onemind agent) |
Interactions and approvals
| Operation | What it does |
|---|---|
| Answer a tool's question or permission | Reply to a pending tool request directly from the onemind agent |
| Approve or decline an assignment request | Decide an AI tool assignment approval |
If the onemind agent's model connection is unavailable, manual task creation and assignment remain available.
Only Team admins can create Projects, invite people, and run Safe Check. When a member asks, onemind explains that an admin must do it instead of preparing a card. An invitation's code and link appear only on your card, never in the conversation. onemind cannot connect an AI tool for you: connecting happens on your Mac.
Action cards
When the onemind agent performs an operation, it returns a structured action card showing the current state of that operation. Review the card and select Run action to execute it, or Cancel to dismiss it. Nothing runs until you select Run action.
Card states:
| State | Meaning |
|---|---|
| Ready to run | The action is prepared and waiting for your confirmation |
| Working | The action is running |
| Completed | The action finished successfully |
| Failed | The action encountered an error |
| Stopped | The action was cancelled |
| Interrupted | The action was cut short; check the result before repeating |
The card updates as the operation progresses. For proposal-bound operations such as Propose a Mind update, the card shows the pending approval state until an admin decides.
Generate Minds with the onemind agent
You can ask the onemind agent to run the Mind Generator instead of using the CLI. The run itself happens on your own device: the onemind agent asks a device where the onemind service is running and Mind generation is turned on (onemind analyze on).
If no such device is available, the onemind agent reports the reason and shows the commands to fix it:
| Reason | What it means |
|---|---|
| No enabled device | No device has Mind generation turned on. Run onemind analyze on, then onemind analyze now. |
| Provider not configured | The model provider for Mind generation is not set up on the enabled device. |
| Run in progress | A Mind generation run is already in progress on your device. |
| Recently requested | A run was triggered recently; wait for it to finish before requesting another. |
- In the onemind agent conversation, type
/and select Run the Mind Generator now. - The onemind agent asks your enabled device to analyze its unread work history.
- Check status by selecting Check the Mind Generator. A draft exists only after a run has completed.
- Generated drafts appear as pending Mind proposals that an admin must approve before they are published to shared memory.
Proposals generated this way appear in the Notification inbox. See Sync and share team memory for Automation mode and the CLI-based approach.
For the full list of BETA plugin workflows involving the onemind agent, see BETA plugin workflows.