Cedarline Home Services
Multi-channel AI agent suiteFour channels, one set of operational facts underneath.
Cedarline Home Services is a fictional business, created by Valtair to show how the system works.
At a glance
- Type
- Demonstration build, publicly reachable
- Sector
- Home services (multi-trade contractor)
- Channels
- Web chat, WhatsApp, voice, email
- Scope modelled
- 7 trades, around 50 services, 12 cities
- Grounding
- Zip-level coverage matrix, schedule, and policy records
- AI approach
- One agent definition with tool access across four channels
- Voice
- Streaming speech with barge-in over the shared tools
- Oversight
- Provisional booking only, confirmed by a callback
- Model providers
- Multiple, provider-agnostic
- Deployment
- Cloud, per-channel session state
- Business
- Fictional, built to demonstrate the system
Cedarline Home Services is a fictional multi-trade contractor in the North Carolina Triangle, and the most demanding build of the two. A home services agent has to answer questions with real operational consequences: whether a crew comes to that address, what the visit costs, whether the failure counts as an emergency at 2am on Thanksgiving. Getting those right is not a model problem. It is a data problem, and the site was built operations-first for exactly that reason.
The scenario
Seven trades, eleven crews, and around fifty services across twelve Triangle cities, where coverage is not uniform: some zip codes get every trade, some get five with a travel fee, some are emergency-only, and some are outside the area entirely. Office hours and dispatch hours are different windows. Holidays override the extended summer schedule. Most of the calls a contractor takes are the same dozen questions about that ruleset, asked at hours when nobody is at the desk, and every one of them is a booking that either happens or does not.
What it demonstrates
Web chat that answers on the operational rules
Coverage by zip rather than by city, the two-hour arrival windows, the $89 diagnostic and when it is waived, what the twelve-month warranty covers. The assistant answers from the coverage matrix and the schedule, so a partial-coverage zip returns the trades that are available and the travel fee, not an optimistic yes.
WhatsApp as a first-class channel
The same agent on the channel customers reach for when something is leaking, where a photograph of the panel or the pipe is worth more than a paragraph describing it. The conversation and its context carry across from the site rather than starting again.
A voice agent that answers the phone and books
It takes the call, works out which trade the problem belongs to, checks coverage against the address, offers arrival windows from the dispatch board, and books one. Emergency calls are triaged against the covered-work list and routed rather than scheduled.
Email automation around the booking
Confirmations, arrival-window changes, and estimate follow-ups, drafted against the job record so the detail in the message matches the job on the board.
One ruleset behind all four
Coverage, hours, holiday precedence, emergency policy, pricing, and warranty terms are held once and read by every channel. The reason a customer gets the same answer on the phone as on WhatsApp is that neither of them is holding its own copy.
System architecture
AI and data components
- One agent definition shared by every channel, with channel-specific presentation
- Tool calls for coverage lookup, availability, booking, and pricing
- Trade classification from a description of the symptom
- Emergency triage against the covered-work and never-emergency lists
- Speech recognition and synthesis with barge-in handling on the voice path
- Drafting of transactional email against the job record
- Provider-agnostic model orchestration
Backend and integrations
- Operations model holding coverage, hours, holiday precedence, and emergency policy
- Zip-level coverage matrix distinguishing full, partial, emergency-only, and outside the area
- Booking and dispatch with two-hour arrival windows
- Estimate requests for work that is quoted rather than diagnosed
- WhatsApp Business, telephony, and transactional email integrations
- Session state carried across channels for one customer
Engineering decisions
The operational data was built before the agents
Coverage by zip, dispatch versus office hours, holiday precedence over the seasonal schedule, the never-emergency list. An agent cannot reason its way to whether Clayton gets carpentry. Most of the accuracy in this build is in the records, not the prompt, which is the general case and not a quirk of this domain.
One agent definition, four presentations
Maintaining a chatbot, a WhatsApp bot, and a voice agent separately guarantees three different answers to the same question within a month. Behaviour is defined once; each channel differs in how it renders and how it handles interruption, not in what it believes.
Nothing is confirmed without a human
Every channel books into a provisional slot and the confirmation comes from a callback. Dispatching a crew is expensive and irreversible, so the agent goes as far as the last reversible step and stops.
Voice was designed around latency, not transcription quality
A caller tolerates a wrong word far better than a two-second silence. The voice path streams, handles barge-in, and keeps its tool calls off the critical path where it can, because the failure mode that ends calls is the pause.
Emergency triage refuses as often as it accepts
Remodeling, cabinetry, lawn, patio, fencing, and cosmetic work are never emergencies, and the agent says so rather than dispatching a crew at a $189 call-out. A triage step that cannot decline is a routing step with extra latency.
What this demo does not do
- The company, its crews, licences, reviews, projects, and prices are fictional. The licence numbers and the yard address are invented and describe no real business.
- No crew is dispatched and no visit happens. Bookings demonstrate the flow through to the provisional slot and stop there.
- The agent channels are the subject of this build; the surrounding site is a demonstration of the operational depth they read from, not a live business.
- Payments, invoicing, and the financing flow are represented in the interface but not connected to a processor.
Voice AI
For inbound and outbound conversational systems with real telephony and CRM integration.
Explore capabilityAgents and Automation
For multi-step workflows that require tool use, decision logic, and human approval.
Explore capabilityRAG and Knowledge Systems
For products that must retrieve and reason over private or domain-specific information with citations.
Explore capabilityBackend and Platform Engineering
For AI systems that need reliable, multi-tenant infrastructure behind them.
Explore capabilityWant one of these for your business?
This was built to be used, not watched. Open it, try to break it, then tell us what you are building and we will show you how we would approach it.