chat command connects to an OAuth-enabled shared Harness through an alias published by the tenant login discovery document. It is intended for many users sharing one Harness Runtime without sharing chat history: users provide only a trusted alias, and the CLI refreshes tenant discovery, resolves the Runtime endpoint, and obtains a fresh id_token from the current OIDC session for each message.
Before using this command, log in with the tenant login address in identity-only mode. The shared Harness must use
custom_jwt authentication and be published by an administrator in shared_harnesses on the same tenant login discovery document.chat
Chat with a shared Harness once or in an interactive session. Passing[message] sends one message; omitting [message] starts a prompt where /exit or /quit ends the session.
Shared Aliases
Shared Harness aliases are resolved only from the tenant login control plane stored in the active profile. The CLI re-reads that control plane’s/.well-known/agentkit-cli document when connecting and before each message, and requires the issuer, OAuth client, and control-plane address to still match the active profile.
Administrators publish non-secret bindings in the discovery document:
Session Isolation
chat does not accept caller-selected Runtime IDs, endpoints, user_id, model overrides, or tool overrides. The server scopes sessions to the verified OIDC user identity; two users may pass the same visible --session-id without sharing a server-side session.
Check connections and sessions
A successful one-message call displays the agent’s response. Reuse your own--session-id in later commands to continue context; omitting it generates a new session
For an unknown alias, check agentkit auth profile show and the alias supplied by your administrator. Sign in again through the tenant address after discovery changes or session expiry. Do not replace an alias with an arbitrary endpoint. Identity-only login supports this chat flow but grants no deployment or resource-management permissions