Conversation behavior and permissions
Routing, follow-up input, quiet hours, approvals and background work.
Documentation for Nookins 0.42.1-alpha.1 · Public alpha
Routes decide when a Nookin answers
A route selects an account, conversation, activation mode and permitted senders/roles. commands handles commands; addressed responds to supported mentions, replies and reviewed hotwords; all_messages handles eligible messages in that exact conversation. Use the route reference before adding groups.
Agent display names are part of addressed activation. Optional literal/regex hotwords are bounded and validated. Voice activation follows the original message's mention/reply/caption, not words discovered later in a transcript. Ambient transcription does not automatically start a turn.
Follow-up messages
Use /mode queue to queue ordinary follow-ups or /mode steer to steer the exact active turn where supported. Media and completion races can fall back to the durable queue. A command changing model/effort does not rewrite an already-running turn.
/new starts a fresh thread. /stop stops the active turn. /task PROMPT starts background work; inspect or cancel it by its handle. See chat commands for the complete registry. Cancellation cannot undo an external action that already completed.
Roles and grants
Owners, trusted senders and members are distinct roles. Routing access is separate from tool authority. Agent features, workspace capabilities, web policy and extension attachments determine what can be attempted. Audience defaults to owner_only; broadening it to agent_callers needs a deliberate review.
Every configured owner is a fleet custodian in a supported private primary-agent session. The primary-owner designation selects defaults; it does not make other configured owners less trusted. Add only mutually trusted owners. Groups and automated/delegated work do not inherit private custodian authority.
A proposal is inert until the required separate human approval. The approval is bound to the exact change and configuration revision; changed source or a concurrent edit can invalidate it. The model cannot approve its own proposal. Clarification and polls are not substitutes for an approval gate.
Quiet hours and standing permissions
Agent interaction.quiet_hours contains an IANA timezone and windows with days, start/end times and allowed roles. Empty allowed roles silence the agent in that window. Overlapping windows intersect permissions. Custom roles describe reply-policy membership, not global tool grants. See interaction fields.
Quiet hours apply at execution and delivery. Suppressed requests/results do not automatically replay in the morning. Cancellation and approval decisions can still be processed. Existing Sentinel quiet-time restrictions remain separate.
private_records_without_confirmation permits only the documented creator-only Records changes for the authenticated requester. It does not authorize arbitrary writes. Personal-calendar standing permission requires exact selected calendar resources and verified eligibility; other calendar writes retain confirmation.
Schedules and Sentinel
A schedule is explicit timed work; an event automation reacts to a registered event. Review timing, destination, agent and grants before enabling either. Inspect schedules through /nookins schedule list or operations CLI.
Sentinel controls scoped attention, learning and participation. Start disabled or summoned. Greater attention does not grant tools or permission for external effects. Use /sentinel help and Sentinel status to inspect qualification and readiness before enabling ambient behavior.