Inbox & conversations

Every channel feeds the same inbox. A conversation is a per-workspace numbered thread between one contact and your team, living in one inbox (channel instance).

Conversations

• Status: open (human-owned) · pending (owned by an AI agent) · snoozed · resolved. Taking over a pending conversation means opening it — the AI agent only replies while the status is pending.
• Priority: none | low | medium | high | urgent.
• Trackers: waiting_since (set on inbound messages, cleared on your reply), first_reply_at, resolved_at, last_activity_at (list sort key). Unread state comes from per-side last-seen timestamps, not counters.

Messages have a visibility:

• public — delivered to the contact over their channel. Outbound public messages carry a delivery status (pending | sent | failed) with the error text on failure.
• note — internal, never delivered or shown to the contact. Use notes for context, @-discussion and AI agent summaries.
• activity — system timeline entries (“Sam resolved the conversation”).

Message content is markdown; attachments are stored server-side and linked from the message.

Assignment and teams

Conversations can be assigned to a member (assignee) and/or a team. Teams are named groups of members (with an emoji and description) — assign a conversation to “Support” and anyone on that team picks it up. Widget inboxes have an auto_assign config flag; it is off by default so new conversations land in Unassigned for the team to triage.

Channels

An inbox is one channel instance; a workspace can have many. Channel types: widget, email, slack, telegram, whatsapp, messenger, instagram, line, sms, api. Each inbox carries its own config (widget theming, email transport, …) and encrypted channel credentials. See Chat widget and Integrations for per-channel setup.

SLAs

An SLA policy defines up to three thresholds in minutes: first response, next response (per waiting episode) and resolution. Optionally count only business hours (only_during_business_hours) so nights and weekends don’t breach.

Apply a policy per conversation (PUT /conversations/{id}/sla) or automatically: set sla_policy_id in an inbox’s config and every new conversation in that inbox gets it. A scheduled scan (every 60 s) records breach events (frt, nrt, rt); the applied SLA walks active → hit | missed | active_with_misses. The SLA report shows attainment per policy.

Automation rules

A rule subscribes to one event — conversation.created, message.created, conversation.status_changed, csat.submitted or contact.created — checks AND-ed conditions, and runs ordered actions:

{
  "name": "Route billing to the billing team",
  "event": "conversation.created",
  "conditions": [
    { "field": "channel_type", "op": "eq", "value": "email" },
    { "field": "content", "op": "contains", "value": "invoice" }
  ],
  "actions": [
    { "type": "assign_team", "params": { "team_id": "…" } },
    { "type": "set_priority", "params": { "priority": "high" } }
  ]
}

Condition fields: inbox_id, channel_type, status, priority, subject, content, contact.email, tag, plus contact.attributes.*. Operators: eq, neq, contains, in, exists. An empty condition list matches everything.

Actions: assign_user, assign_team, set_priority, set_status, add_tag, send_reply, send_note, notify_member, send_webhook. Rules run in order (ord) and can be disabled without deleting.

Macros

Macros are the same action vocabulary (plus remove_tag), saved as a named shortcut a member runs manually on a conversation — “tag as bug, reply with the triage template, assign to engineering” in one click. Visibility is personal (creator only) or global (whole workspace). Reply/note content supports {{contact.name}} and {{agent.name}} placeholders.

Reporting

GET /api/v1/w/{workspace_id}/reports/overview?days=7 returns:

• Volume: new and resolved conversations, per day and per channel.
• Resolution: resolution rate, median first-response and median resolution minutes, resolved-per-agent breakdown.
• CSAT: average rating and response count. When a conversation is resolved, the widget asks the contact for a 1–5 rating with optional feedback.
• AI: agent runs, AI-resolved count and AI resolution rate.

A breakdown endpoint slices the same numbers by inbox, team, agent or tag, and every report has a CSV export variant. SLA attainment lives at GET /reports/sla.