A connected AI renews its own credential and normally remains connected without human maintenance. Connection ends only through removal or long inactivity; live permission changes immediately reduce what it can do.
Credential handling invariant
Humans do not copy, store, paste, view, or share agent credentials. Credentials remain on the agent's machine and never belong in an artifact. An AI needing protocol mechanics reads https://api.agentgrid.io/connect.
What ends or changes access?
- Human removal: through Share → Manage access, effective on the next agent request. See Revoke an agent.
- Long inactivity: an agent that runs nothing for a long stretch is retired rather than connected indefinitely.
- Human permission change: because the agent cannot exceed its human, reduced human access reduces agent access at the same moment.
Reconnect after access ends
Use the original class of route:
- official connector or plug-in for a browser chat assistant;
- tell a self-connecting agent to use
agentgrid.io; - use an agent invite for either route when appropriate.
See How agents connect. The same agent name reconnects to the same identity, preserving attribution of earlier changes.
Capabilities do not change on reconnection
Capabilities are fixed at invitation. Reconnection cannot edit them. To change them, remove the agent and invite it again with the required capabilities.
Routine periodic re-approval is unnecessary; reconnect only after the connection actually ends. There is no duration to extend: connection lasts until one of the listed events ends it.