Global agents for organization-wide AI.
Global agents let you create AI agents once at the organization level and make them available across all projects. They are ideal for shared roles like code review, security auditing, or onboarding guidance where you want consistent behavior everywhere.
How to set up and manage global agents
Global agents are configured from your organization settings:
- Open Settings, then Agents.
- Use the Agents sidebar to list all existing global agents.
- To add one, click New Agent and choose:
- Create with AI: describe what the agent should do and the AI will generate its instructions. - Create Manually: enter the agent name, slug, description, and system prompt in Markdown.
- Select an agent to open the Agent editor and revise its system prompt, description, or other details.
- Only organization owners can create, edit, or delete global agents.
Global agents show up automatically in each project's agent list with a Global indicator. Project teams can use them but cannot delete them; only organization-level administrators manage global agents.
When you delete a global agent, it is removed from all projects, so use deletion sparingly.
How to decide between global and project-level agents
Use global agents when:
- The role is shared across projects (for example, "Senior React reviewer," "Compliance checker").
- You want all teams to follow the same standards.
- You want a single place to update instructions.
Use project-level agents when:
- The expertise is specific to one project's stack, codebase, or domain.
- Only a subset of teams needs the role.
Global agents work alongside the default Workspace Assistant. The Workspace Assistant is the general AI helper available everywhere; global agents are specialized personas you define for particular kinds of work. Organization settings include a dedicated Workspace Assistant panel where you can:
- Enable or disable the assistant for the whole workspace.
- Edit its system prompt in Markdown.
- Control which MCP servers and tools the assistant is allowed to use.
Use global agents for patterns that span projects: code review standards, security checks, documentation quality, or release notes. Create one agent per stable role, not per one-off task. Review global agents after a project or policy change, because a once-correct agent can quickly become out of date. Pair global agents with project skills for project-specific constraints; let the global agent define the pattern and the local skill enforce the local rule. If a global agent is rarely used, rename or narrow its instructions instead of leaving it as a vague “assistant” for everything.
Treat global agents like long-term employees: they should have clear responsibilities, consistent instructions, and a review cadence. If a global agent is used across many projects, audit its instructions after major policy changes. Keep global agent behavior aligned with organization norms, not a single team’s habit. If you notice conflicting instructions between global agents and project-specific skills, fix the conflict before expanding usage. That keeps AI behavior predictable and safe.
Related entries
- project-skills-global-skills
- coding-agent-design-engine
- invite-manage-organization-members