Every reservation you don't confirm is a table you don't fill.

PIPPA · PIPELINE Limited Rollout

Lead Flow & Front Desk

PhoneVoice

A full Telnyx provider layer (lib/telephony/), an idempotent provisioning state machine that reconciles rather than re-orders a paid number, tenant-safe phone-to-slug lookup, and a per-slug inbound voice webhook with idempotent call logging and per-slug Slack routing all now exist and are unit-tested — including the no-credential and no-real-call paths. None of it has run against a live account: no Telnyx account exists, no phone number has ever been ordered, and no real inbound call has ever reached the webhook.

1. 1

   Voice agent profile provisioned per business — greeting, hours, and FAQs built into its script

2. 2

   Call outcomes log automatically with transcript, duration, and outcome — matched to an existing lead by phone number — once a real phone account exists; none is today

CHASE · LOCKDOWN Limited Rollout

Sales & Deal Closure

EmailCRM

The real follow-up mechanism is the hub's generic FollowUpTask queue (cadence, opt-out, reply-cancels-chase) — not the 'follow-up-nurture' n8n AI-drafting workflow, which is inactive with 0 executions and templated to expect deal-context fields nothing in the codebase ever populates. No lead scoring or CRM pipeline-stage automation exists anywhere in the chase path.

1. 1

   Follow-up tasks queue automatically and drain on a schedule — replies cancel every pending message

2. 2

   Per-client send cadence honored — your own follow-up pace, not a hardcoded default

3. 3

   Opt-out is enforced permanently — STOP or unsubscribe and that contact is never messaged again

CLEO · CROWN Limited Rollout

Cold Outreach & Email Campaigns

Real 3-step campaign sequencing (lib/outreach/{sequence,cleo}.ts), per-prospect personalization from structured caller-supplied data, per-client suppression/opt-out, and a hard CAN-SPAM postal-address gate are built and reachable via api/outreach/cleo/{enroll,run,suppress} — the same two-phase atomic-claim discipline RICO and NOVA use. The catch: every delivered client sends through one hub-wide RESEND_API_KEY sending domain, so one client's cold-list bounce/complaint rate would degrade deliverability for every other client's transactional mail too. CLEO is safe and honest to run for Inteliworks' own outbound prospecting only — not as a delivered client's own branded campaign tool — until per-client sending-domain verification exists. Reply-cancels-sequence is wired at the site GOAL.md named (any reply through an existing channel cancels remaining steps), but no inbound email-parsing webhook exists anywhere in this codebase, so a prospect literally hitting 'Reply' in their inbox never reaches it. No provider bounce webhook exists either, so hard-bounce suppression is manual, not automatic. Also resolves a real drift bug this run found: apps/hub/src/lib/personas.ts previously defined CLEO as 'Client Experience & Retention' (onboarding-automator/review-collector/customer-support — NOVA/GLORIA/VERA's jobs), contradicting this fleet.ts definition that the built code actually matches; personas.ts's display copy (role/tagline/hook) was corrected to match, its provisioning-facing coreAgents left untouched as a separate, lower-risk follow-up.

1. 1

   A real 3-step sequence — intro, day-4 follow-up, day-10 follow-up — runs and sends automatically once enrolled

2. 2

   Personalizes from the company, contact, and fact data you supply — never invented, and refuses to enroll anyone missing a real name, company, or email

3. 3

   CAN-SPAM handled: a real postal address is required before anything sends, every email carries one-click unsubscribe, and suppression is enforced on every send

4. 4

   Running today for Inteliworks' own prospecting only — every delivered client currently shares one sending domain, so client-branded campaigns wait until each client has their own verified domain

RICO · HUSTLE Limited Rollout

Revenue Recovery & AR Automation

Email

Real escalation-ladder dunning is built and reachable (lib/dunning/{ladder,rico}.ts, model InvoiceDunningTask): reads open invoices straight from Stripe, applies the friendly/firm/final ladder, includes Stripe's own hosted_invoice_url, records what was sent with a two-phase atomic claim so nothing is dunned twice, and re-verifies payment status immediately before send. The catch: there is only ONE Stripe account in this codebase — no Stripe Connect, no per-delivered-client billing integration — so RICO dunns Inteliworks' OWN delivered clients for THEIR overdue retainer invoices TO Inteliworks. It does not and cannot dun a delivered client's own downstream customers (a med spa's patients, a restaurant's diners); those live in billing systems this codebase has no visibility into. fleet.ts's copy is corrected this run to describe that real scope instead of implying client-facing AR automation.

1. 1

   Escalates automatically — friendly, firm, then final — the moment each threshold is crossed, and never sends the same step twice

2. 2

   Every reminder carries a real, working Stripe payment link — never a placeholder

3. 3

   Stops the instant an invoice is paid, re-checked again right before it sends

4. 4

   Running today on Inteliworks' own invoicing — automating this for your own customers' invoices needs your own connected billing account, not wired up yet

QUINN · TRENDING Limited Rollout

Content & Social Media

InstagramFacebookLinkedInX

AI content drafting for 4 platforms is real. A real Postiz publish node exists in the n8n workflow, but zero .ts/.js in the repo references Postiz, the Postiz stack is in neither docker-compose file, and the node has 0 executions. No scheduling trigger exists — the workflow only runs on-demand via webhook. No post-performance feedback loop exists.

1. 1

   Content workflow turns one source piece into platform-native posts

2. 2

   Every post drafted in your brand voice from your services and offers

3. 3

   Drafted posts can auto-publish via Postiz once that integration is deployed — not wired up yet

VERA · CALM Limited Rollout

Customer Support & Care

Web ChatEmail

Ticket creation, keyword triage, and Slack escalation are real and reachable. The claimed 'conversation memory' and 'intent branching' overstate a single boolean keyword regex, and no code exists anywhere to surface recurring-issue patterns across tickets.

1. 1

   Every support request logged as a ticket with a full event trail

2. 2

   Keyword triage flags support requests so they don't get treated as sales leads

3. 3

   Escalated tickets open an internal petition and page the team on Slack

ROMAN · PRESTIGE Limited Rollout

Proposals & Business Writing

The proposal-generator n8n workflow ROMAN actually runs is a Webhook -> chainLlm -> respondToWebhook stub with static Inteliworks-only pricing baked into the prompt — no deal-record lookup, no storage, no delivery. A real proposal storage/versioning/e-signature system exists in the codebase (concierge/proposal.ts, /api/proposals), but it is Inteliworks' own internal sales tool, not keyed to the roman persona or the proposal-generator slug. No PDF generation exists anywhere in apps/hub.

1. 1

   AI drafts a full proposal — summary, recommended plan, pricing, and timeline — from a written brief

DASH · RAPID Limited Rollout

Hot Lead Routing & Rapid Booking

CalendlyEmail

Real, newly-built Google Calendar availability/booking API exists (check_availability, create_meeting) gated behind hasOperator(config,'dash'). No client has dash turned on and no GOOGLE_OAUTH_*/BOOKING_WEBHOOK_SECRET_<SLUG> credentials exist. The separate 'hot-lead-calendly-router' n8n workflow is a quota-gated outreach email with a static Calendly URL embedded as a link — it never calls the Calendly API and has no lead-score gate at all.

1. 1

   Real-time availability and booking against Google Calendar via check-availability and create-meeting

2. 2

   Quota-gated outreach email drafts a personalized invite with your booking link

3. 3

   Booking confirmed and logged before the lead has time to cool

GLORIA · ICONIC Limited Rollout

Reputation & Review Management

Email

Real review-request sending and the satisfaction gate are built and reachable (lib/reviews/{gloria,response-link}.ts, model GloriaReviewTask), triggered off completed BookingAppointment rows: the request fires after every completed appointment, the one-tap link goes out only when the client's own fleet.json supplies review.publicReviewUrl (never guessed, skipped and logged if missing), and unhappy responses are structurally diverted to private feedback — publicReviewUrl is read in exactly one place, inside the happy branch, so no code path can leak a public review link to an unhappy response. What is NOT built, because it genuinely requires a Google Business Profile OAuth connection nobody has: reading back a live review count, rating, or 'velocity' — fleet.ts's old fourth claim ('review velocity tracked so your rating climbs') is dropped this run rather than rewritten, since no proxy metric exists to honestly stand in for it.

1. 1

   Sends automatically after every completed appointment — no list to manage by hand

2. 2

   The one-tap review link is always your real page — never guessed, and skipped entirely if you haven't given us one

3. 3

   Unhappy responses are caught first and routed to a private inbox — never a public review site

NOVA · LAUNCH Limited Rollout

Client Onboarding & Welcome Sequences

Email

Real welcome/onboarding sequencing is built and reachable (lib/onboarding/{sequence,nova}.ts, model NovaOnboardingTask), triggered by a Lead captured through a delivered client's own intake webhook — never Inteliworks' own signup flow. Three real, personalized steps (welcome +0h, tips +48h, stall_nudge +120h) fire from the lead's actual intake answers, and a reply, conversion, or opt-out cancels every remaining step the instant it happens — 'stalled' is honestly defined as none of those three having happened by the time stall_nudge comes due, not a separate inferred signal. Enrollment reuses the exact ChannelContact VERA already resolved for the same lead, which means NOVA enrollment requires VERA to have run for that request; every client today with NOVA configured also has VERA on, so this is a documented design limitation rather than a live gap.

1. 1

   Enrolls automatically off your own site's lead capture — welcome at 0h, tips at 48h, a stall nudge at 120h

2. 2

   Personalized from the lead's own name and their own words from your intake form — never generic filler

3. 3

   A reply, a booking, or an opt-out cancels every remaining step the instant it happens

MILA · CONNECT

WhatsApp & Telegram Outreach

WhatsAppTelegram

1. 1

   Live webhooks on WhatsApp and Telegram — every message lands instantly

2. 2

   Instant in-channel confirmation: they always know MILA got it

3. 3

   Automatic follow-ups until they reply or book — replies cancel the chase

4. 4

   Every thread logged per contact; hot leads route straight to booking

BELLA · VIRAL Limited Rollout

Instagram & Social DM Engagement

InstagramMessenger

Client-scoped Meta webhook and per-client credential plumbing shipped this session, mirroring MILA's real send/confirm/follow-up loop. No client has Meta tokens configured yet, and there is no in-conversation lead scoring anywhere in the codebase.

1. 1

   Meta webhook covers Instagram DMs and Messenger in one thread

2. 2

   Instant confirmation on every DM — no lead waits until morning

3. 3

   Quiet threads get automatic follow-ups; replies cancel the chase

4. 4

   Every thread logged per contact, the same trail MILA uses