Shared Agents
Last updated:
Shared agents let admins and moderators make one configured agent available to other users while still allowing per-user personalization.
Share an agent
Admins and moderators can turn on Share with all users in a custom agent's editor. Leave the allowed-user selection empty to make it available to everyone, or select specific users to restrict it.
Admins and moderators can edit the shared definition. Other users see Customize instead of Edit. Customize opens only the personal context editor. It does not allow changing tools, instructions, model providers, or file access.
Sharing controls visibility, not ownership of every setting. A shared agent still runs with the saved shared definition, and users who cannot manage shared agents cannot delete it or change its runtime surface.
Personal context
Personal context is text stored for one user and one shared agent. Use it for preferred names, output style, team names, recurring repositories, responsibilities, or other stable context.
For personal context to affect responses, the shared agent's system instruction must include {{ user_context }}. Without that variable, the context is saved but not injected.
Note
Personal context is per user. It is not a global edit to the shared agent.
Personal context changes are not saved automatically. Click Save in the Customize view before returning to the agent list.
Do not put secrets, passwords, or API keys in personal context. Use Credentials settings for secrets.
If an administrator removes a user from the shared agent's allowed list, that user's saved personal context for the agent is removed.
Permissions and tools
Sharing an agent does not grant admin rights, but it does make that configured assistant available to the selected audience. Review tools, workspaces, model provider, memory, skills, and exposed agent tools before sharing. The called agent keeps the permissions saved on the shared definition.
Caution
If a shared agent has write-capable tools, connected credentials, or writeable directories, allowed users can ask it to use them within its configured policies.
When to share
Shared agents work well for standardized assistants such as support triage, release notes, operations summaries, or internal knowledge helpers. Keep experimental or personal agents private until their instructions and tool access are stable.
Use user restrictions when the shared agent contains team-specific knowledge or workspace access. Keep credentials in the credential store and grant the narrowest tool access that works.