Sharing and using agents
Make an agent available to selected people or the whole workspace. Each person uses it with their own permissions on brains and accounts.
A new agent is private. Once it works reliably, share it with the people who do the task. Each person uses it with their own permissions on brains and accounts.
Share an agent
Open the agent
Under Agents, choose Edit on the agent and scroll to Sharing.
Choose the visibility
Choose a level under Visibility (see the table). With Selected people, search under People with access; with groups (on request), also choose under Groups with access.

Save
Click Save changes. The people find the agent on the Shared tab.
| Visibility | Who sees and uses the agent |
|---|---|
| Private (only me) | only you, not even admins |
| Selected people | you, the chosen people and all admins; with groups, all current members of the chosen groups. Without anyone chosen, the agent is saved as private. |
| Shared with workspace | all members |
In a workspace with groups, your admin decides who may share with everyone. Otherwise Shared with workspace is locked with the note "Your groups can't share agents with everyone. Ask an admin." (Groups).
Use an agent
Start a chat under Agents with Start chat, or type @ in a new chat and choose under Agents. The top of the chat shows "Agent:" with the name. The empty chat shows the conversation starters, and the agent's files are already in the message box.

- Your chats are yours: the person who created the agent does not see them.
- You change model and brains in the chat as usual.
- The agent uses connectors through your own connection; if one is not connected, it does not use it for you.
What applies after saving
Instructions, web search and connectors apply from the next message in all chats with the agent. Model, brains and files apply to newly started chats.
- Access lost: if sharing is narrowed or you leave a group, your chats stay. They then answer like ordinary chats, without the agent's instructions and without its files; your own attachments stay.
- Agent deleted: the chats stay and keep the last saved instructions.
- Brains: each person only gets answers from brains they may read themselves. If they lack access, the empty chat says so without naming the brain.
Who may do what
| Action | Creator | Admins | People with access |
|---|---|---|---|
| Start a chat | yes | yes, except others' private agents | yes |
| Edit, duplicate, Delete agent | yes | yes, except others' private agents | no |
| Read others' chats with the agent | no | no | no |
If the person who created a shared agent leaves the workspace, it passes to an admin; private agents stay with them and are visible to nobody. Under My shares you end a share with Stop sharing.
Best practices
- Start small: two or three people first, then the whole workspace.
- One person in charge, who maintains the agent and collects feedback.
- Test first, then save. Saved instructions reach all ongoing chats at once; the History helps you go back.
Frequently asked questions
You may not read one of its brains; the empty chat says so. Ask the person responsible for the brain for access.
Someone saved new instructions, which apply to ongoing chats too. Or your budget is used up: then the economy model answers and tools pause (Budgets and limits).