Features
Every module below ships with tests and is enforced by the API, not a roadmap card. The numbers are counted from the codebase, not from a marketing deck.
Modules shipped
8
every one in the installer today
Permission keys
97
enforced in UI + API, 5 system roles
Automation triggers
13
dry-run before any rule goes live
Backend tests
1,000+
run on every change to the codebase
Channel readiness varies
Inbox surfaces cover WhatsApp, Instagram DM, and Facebook Page conversations. Production connection and reply readiness varies by channel; check pricing for the current state. Assignment, internal notes, and customer history stay beside each thread.
Real CRM records
Profiles with identities across channels, notes, tags, activity trails, archive/restore, and phone deduplication. Search covers conversations and records together.
Back office
Orders with line items and payment status, products with variants and stock movements, and catalogues with featured positioning. All are linked to the conversations that generated them.
Live delivery gated
Approved WhatsApp templates, audience freezing, and immutable campaign snapshots. Durable job queues support retries and pause/resume; live delivery remains gated and is not enabled today.
TKT-YYYY-NNNNNN
Ticketing with timelines, priorities, and assignment. Tasks with lifecycle actions. Both cross-linked to customers, orders, and broadcasts.
13 triggers
A durable trigger-event queue uses an allowlisted condition evaluator, with no eval or template injection. Every rule runs in a savepoint, and every execution is logged.
Draft-only
Suggested replies drafted from the customer's real history. Structurally incapable of sending: the executor re-refuses send actions and automatic-send startup aborts.
Real data
Overview metrics, conversation, ticket, and broadcast time series, distributions, and agent metrics are computed from your actual data, not sampled marketing math.
How a conversation becomes a paid order
On WhatsApp, Instagram, or Messenger. The webhook is signature-verified and deduplicated before anything touches your data.
The conversation lands in the shared queue with the customer's full CRM record beside it: orders, tickets, history, and notes.
Ask for a suggested reply. The AI drafts from real history; nothing is ever sent by the AI itself. You edit or rewrite every word.
Create or update the order from the conversation. Changes stay attached to the order record and its status.
Review changes in the audit trail and provider attempt log, including what happened, by whom, and when.
Counted, not claimed
1,000+
Every change to the backend runs the full suite against a real database before it can merge. Counted from the test directory, not from a slide.
Permission keys
97
enforced in UI + API, 5 system roles
13
Triggers
Durable workers
broadcasts and automations run on leased queues
Trust architecture
Five system roles plus custom roles. Every action button, route, and API dependency is gated in the UI and re-checked server-side.
Every mutation of consequence writes an immutable audit event: customer changes, order updates, broadcast sends, ticket lifecycle, permission edits.
Channel and payment credentials are encrypted at rest, with a separate key for each type of secret. Access tokens live in memory only, never localStorage or disk.
Deployment architecture and a kit exist, but no customer-owned production instance has been provisioned. Confirm hosting and support scope in writing before procurement.
Request access, register a workspace, and put every module above under your own finger in minutes.
