Skip to content

Chat Commands ​

All channels (Telegram + Discord) ​

CommandDescription
/start or /helpGreet and show help
/model <alias|provider/model|model-id>Switch active model
/statusService status — model, last message, coding agents, cron jobs
/cron listList scheduled jobs
/cron run <job-id>Trigger a cron job on demand
/heartbeatTrigger an on-demand heartbeat check
/silence <minutes>Suppress proactive messages for N minutes (default: 30)
/focusEnable focus mode
/approvalsList pending exec approval requests
/approve <id>Approve a pending exec request
/deny <id>Deny a pending exec request
/tasksRedirect to /agents or /cron
/cancel <id>Redirect to dashboard or /cron
/clearClear conversation history
/compactSummarize and compress conversation history

Telegram-only ​

CommandDescription
/memoryList agent memory files
/memory <filename>Read a specific memory file

Discord-specific behavior ​

  • Reasoning effort — /effort <none|low|medium|high|xhigh> or /think <...> changes the current Discord effort override.
  • Reusable agent profiles — /agent manages Discord aliases such as @reviewer; @alias <message> invokes a profile directly. See Agents.
  • Interactive approval cards — When exec approval is needed, Discord sends a message with Approve/Deny buttons (Telegram uses inline keyboards).
  • Threaded replies — Coding agent tasks and Discord agent-profile invocations can create threads from the triggering message. Coding-agent status updates and completion notifications route to the thread. Disable coding-agent thread creation with threadedReplies: false in Discord config.
  • Message chunking — Long responses are split at paragraph/line boundaries into chunks of 1900 characters (Discord's limit is 2000).
  • Image and voice support — Image attachments are analyzed via the model's vision capability. Voice messages are transcribed (Whisper) and optionally replied to with TTS audio.

Maintainers: for Discord posting/workflow changes, follow Discord Update Documentation Process.

Non-command messages ​

Any non-command text is treated as a chat prompt to the agent. Recent conversation history (last 5 pairs) is included automatically.