What a plugin is
A plugin is a capability with a declared scope: a name, description, category, execution sandbox, and permission boundary. When you install a plugin, Spaces adds it to the set of tools it can consider when planning a response. The model decides when a plugin is relevant; you review what it returns. Plugins are not separate chatbots. They do not replace the conversation — the model plans, the plugin executes a specific operation, and the result comes back to the same thread where you can review, revise, and continue.Plugin lifecycle
1
Discover
Open the Plugins screen from the Spaces navigation. Browse featured plugins, explore categories, or search by name or capability.
2
Inspect
Open a plugin detail view to read its purpose, developer, version, execution sandbox type, run count, and rating. Understand what it can access before installing.
3
Install
Click Install. The plugin becomes available in all your conversations. For connected service plugins (GitHub, Notion, Slack, etc.), you will be prompted to authorize access via OAuth.
4
Use
Describe what you want in a conversation. Mention the plugin by name for deterministic routing, or let Spaces select it automatically based on your request. Provide the required inputs — files, links, account connections.
5
Review
Inspect the returned artifact, citations, file, message draft, or action log. Ask Spaces to revise the result before sharing or acting on it.
6
Manage
Use the plugin action menu (⋯) to uninstall, view terms, or disable automatic updates. Removing a plugin does not delete conversation records that referenced it.
How plugin execution works
Every plugin invocation follows the same pattern:- Input collection — Spaces identifies the required text, files, URLs, settings, or credentials for the operation
- Permission check — read, write, send, and account actions are distinguished; a draft stays a draft until you approve it
- Isolated execution — the plugin runs in its declared sandbox (AI Sandbox, Web Browser, Command VM, or Document tool)
- Result validation — output is checked for a usable response, supported format, and actionable errors
- Conversation hand-off — the result and an audit log entry appear in the conversation thread