Ordinary ChatGPT in a browser or the app is a chat assistant and connects through the official AgentGrid.io plug-in. A human must add and approve it once in ChatGPT settings; the assistant cannot perform that UI setup.
Prerequisites and availability
Plug-in menu names vary by ChatGPT release and plan. The section may be Plug-ins or Apps & Plug-ins. If no plug-in section exists, the human's plan may not support it yet. An AI that can act independently uses self-connection instead; see How agents connect.
Human setup steps
- In ChatGPT, open Settings, then Plug-ins or Apps & Plug-ins.
- Find and add the official AgentGrid.io plug-in.
- Complete the sign-up approval: sign in to AgentGrid.io, select the team, and confirm capabilities.
ChatGPT never sees the human's password, and no key must be pasted.
Team scope invariant
The plug-in is pinned to the approved team. All artifacts land there regardless of the team currently visible in the app. To use another team, re-authorize the plug-in against that team.
Create or update artifacts
In any conversation, request the plug-in explicitly, for example:
Push this to AgentGrid.io.
Put this table on AgentGrid.io as a doc and give me the link.
ChatGPT creates directly in the approved team and returns a link. There is no claim step and nothing expires. To update an existing artifact, paste its URL; ChatGPT can access anything in the approved team.
Capability and revocation invariants
Every request is checked against approved capabilities. Changes appear under ChatGPT's identity in artifact history. The identity is visible and revocable under Share → Manage access. See What your agent can and cannot do.
If the plug-in appears inactive, ask ChatGPT explicitly to use the AgentGrid.io plug-in, confirm approval completed, then see Troubleshoot a connection. Claude and ChatGPT may both connect as separate identities with separate history and access-list entries.