Commands

Slash commands are intercepted by MessageGateway before the message reaches the agent. Recognition, channel completion, and the /help text are all derived from a single list, SLASH_COMMANDS in core/src/services/commands/registry.ts, so a command is added or retired in exactly one place.

Reference

CommandUsageWhat it does
/help/help [command]Show help; /help <command> for detail on one command.
/statusConnection, AI provider, model and session mode.
/usage/usage [days|today]Token-usage report from the audit log.
/whoamiHow the agent sees you: channel and access level.
/memoryWhat earlier context has been summarised into this session.
/clearEnd this session and start a fresh, empty one. Nothing is carried forward.
/compactSummarise this session into memory, then start a fresh one seeded with the summary.
/allow/allow <domain>Add a domain to allowed_domains. Trusted senders only.
/exitHow to leave the session. Listed on the TUI only.

/reset is an alias for /clear. /quit and /bye are aliases for /exit.

Notes

  • /usage aggregates LLM calls, tool calls, tokens and wall-clock time. /usage is all-time, /usage today is since midnight, and /usage 7 is the last 7 days.
  • /clear vs /compact/clear drops the current thread entirely; /compact keeps a summary. See Concepts for how sessions and summarisation work.
  • /allow edits the domain allowlist that gates outbound requests. Only trusted senders may run it. See Security.
  • /status reports which provider and model are serving the current session. See AI providers.

Access

Commands marked trusted are limited to authorised users. An unauthorised sender who runs one is told to ask the administrator to add them to the allowed list.