Connect the bot
Create a bot through BotFather and store its token in the private root.env under TELEGRAM_TOKEN. Keep an existing .env intact when adding the variable.
Add the platform to the Entity declaration:
/start. Use run or worker; talk runs only the terminal conversation.
Establish your identity and access
Send/whoami to obtain your Telegram user ID. Use that same canonical person ID when pairing a spatial shell to the relationship.
Configure NGRAM_TELEGRAM_ALLOWED_USER_IDS to restrict the intended audience and NGRAM_TELEGRAM_OPERATOR_IDS for operator controls. These can also be configured in the platform’s allowed_user_ids and operator_user_ids lists. Keep personal identifiers in your private deployment configuration.
Useful commands
Automatic compaction reports progress in the chat. Manual compaction runs without blocking incoming control updates, so an operator can pause while a summary is being generated.
Body command
/body reads the body representation rather than asking the model to invent a state report. Treat it as simulated runtime state, not a measurement of subjective experience.
Busy indicator
/busy on, /busy off, and /busy status control the chat’s busy indicator. /wakequiet on suppresses visible wake activity. These settings affect presentation; they do not pause model use.
Media and groups
The adapter can pass supported image and audio inputs to the harness. Actual perception depends on the model and media-processing configuration. Test those paths separately from text chat. In groups, response routing depends on mentions, replies, and configured behavior. Review bot permissions and access settings before inviting the Entity into a shared conversation. The direct/pause command is the inference stop control. /session_stop belongs to remote desktop sessions and does not stop the Entity worker.