PROTOTYPE A homeowner calls after hours. Nobody picks up — watch what happens next.

Good morning, Maria

Tuesday · 2 crews out · everything routine is running quietly
● All systems normal
Booked this week
$4,310
▲ 18% vs last wk
Avg. lead response
31 sec
industry avg: 4 hrs
Outstanding invoices
$1,240
2 overdue
Jobs today
7
2 crews · on schedule

Needs your attention exception-first: routine work stays quiet

!
New lead waiting — Sara K. wants a deep clean Thursday
Missed call 8:42 PM yesterday · AI texted back in 31s, gathered details, held a slot
A5 · your call
$
Invoice #1041 — 14 days overdue ($380)
Reminder drafted. Sending it is your decision, not the AI's.
A3 · drafted
RT
Rosa's cleaning certification expires in 12 days
She'll drop out of eligible-crew lists automatically if it lapses.
A1 · heads-up

Morning briefing ✦ AI

Assistant · every fact cites a record
Cash on hand covers 5.2 weeks of payroll. Thursday has a 2-hour gap on Crew B — the new lead from Sara K. fits it exactly. Two recurring clients are due for their quarterly deep-clean upsell. No actions were taken.

Sara K. · (415) 555-0148 new leadsource: Google Business Profile

Inbound call · 8:42 PM · no answer
Missed call — after hours.
✦ AI · 8:42 PM · 31 seconds later
Hi! This is Brightside Cleaning — sorry we missed you. How can we help? (Reply STOP to opt out)
Sara · 8:47 PM
Hi! Need a deep clean for a 3bed/2bath in Noe Valley before family visits Thursday.
✦ AI
We can do that. Deep cleans for 3bd/2ba typically run $280–$340 depending on condition. Thursday morning looks open — want me to have Maria confirm a time and exact quote first thing tomorrow?
Sara
Yes please! Morning works.
✓ replied in 31s slot held · expires 10 AM A2: quoted a range, never a price

Why this matters

62% of calls to small service businesses go unanswered
Each missed call is a customer choosing a competitor. The AI answers every channel, 24/7 — and hands the decision back to the owner.
The AI never invents availability or prices
It offers ranges from the owner's price book and holds tentative slots only from the real calendar. Deterministic systems decide; the AI explains.
Compliance built in
STOP/HELP handling, quiet hours, consent records — carrier and TCPA rules enforced by the platform, not by the owner's memory.

Thursday, June 18

Same engine, your words: visits, crews, service plans.

Crew A — Rosa & Tina
Crew B — Marcus
Crew C — open hire
8 AM
Weekly · Hayes St2 hrs · recurring plan
Move-out · Dolores3 hrs · one-time
↓ 14 min drive
↓ 9 min drive
11 AM
Bi-weekly · Castro2 hrs · recurring plan
+ Sara K. — deep cleanheld from lead · fits the gap
awaiting your confirm
2 PM
Weekly · Noe Valley2 hrs · recurring plan
Quarterly deep · Missionupsell opportunity flagged

One schedule engine serves every trade: dispatch calls (HVAC), recurring routes (cleaning), project milestones (roofing). Recurrence follows the iCalendar standard — skip one visit without breaking the series.

Quote #Q-118 — Sara K. draft

ServiceQtyPrice
Deep clean — 3 bed / 2 bath1$300.00
Inside oven & fridge add-on1$45.00
Tax (SF county · provider-verified)$0.00
Total$345.00
✦ AI drafted the words — never the numbers "Our two-person crew will deep-clean every room top to bottom, including baseboards, inside windows, oven and fridge — your home guest-ready before Thursday evening."

What the customer sees

No account, no app, no login
A secure link by text. One tap to approve, typed signature, done. The approval binds to this exact document version — change a line, and re-approval is required.
Prices come from the owner's price book
Totals are computed server-side in exact cents. AI literally has no write access to monetary fields.
Approved quote → job → invoice, automatically linked
No re-typing, no lost paperwork. The whole chain is auditable end to end.

Needs you AI proposes · deterministic policy verifies · you decide

A5 · consequential — human required proposed by ✦ Assistant

Confirm booking & send quote #Q-118 to Sara K.

Books Crew B, Thursday 11 AM – 2 PM, and sends the $345.00 quote for approval. The held slot expires at 10:00 AM if you decline.

Exact amount
$345.00 USD
Document version
Q-118 · v2 · #a41c
Approval expires
Today 10:00 AM
Reversible?
Yes, until customer approves

✓ Approved by Maria (Owner) · single-use token · logged forever

Quote sent to Sara by text. Slot confirmed on Crew B. If Sara approves, the job, checklist and invoice are created automatically — and her card mandate enables auto-pay on completion.

What AI can never do here — by architecture, not policy text
Charge a card · change a live price · file taxes · submit payroll · accept a contract · hire or fire. These are A6: there is no code path. That's the trust story we sell to small businesses.

Invoices money is integer cents · every total reproducible

#1043 · Hayes St weekly — $120.00
Auto-paid on completion · customer mandate on file · tax evidence cached
paid
#1042 · Dolores move-out — $410.00 +$25 tip
Paid by link · Stripe-hosted · platform never touches card data
paid
!
#1041 · Mission quarterly — $380.00
14 days overdue · reminder drafted, awaiting your send
overdue

This month

Collected
$11,860
Job profitability (directional)
38%
2 jobs missing labor cost — flagged, never guessed
Recurring revenue share
64%
✦ Business Assistant answers cite records · no actions taken
✦ Assistant
Hi Maria — ask me anything about the business: money, leads, schedule, team, taxes, reviews, growth. Every answer comes from your records, with sources. I can explain and propose; I never act without you.

One platform · optional intelligence systems each pack reads governed data — none owns the truth, none acts alone

Small businesses turn packs on only when useful, inside the same flat price and AI budget. Every pack obeys the same rule: it can observe, explain and propose — consequential action always returns to the owner.

NexusOS active

The business assistant. Cross-domain briefings, "can I afford it?" answers, scenarios — every fact cites a record.

Boundary: never executes · owns briefings, not source data
ReviewArmor active

Asks every customer for honest feedback after the job, tracks reviews and complaints, drafts recovery responses.

Boundary: no review-gating · publishing needs the owner
AuditMax active

Operational finance intelligence: cash runway, receivables, job profitability, payout reconciliation, finance exceptions.

Boundary: no money movement · all math deterministic
Tax Intelligence optional

Tax readiness and evidence: jurisdiction mapping, gaps, exemption issues, professional-review packets.

Boundary: never files or pays · provider-verified rates only
VibeLocal optional

Growth intelligence from first-party truth: which leads convert, where demand comes from, what customers actually say.

Boundary: no automatic ad spend · scenarios, not campaigns
GeoPulse optional

Outside-world research: territory demand, permits, labor, weather, suppliers — from official sources, with provenance and cost shown.

Boundary: paid lookups need approval · estimates never overwrite facts
TalentForce optional

Workforce intelligence: capacity, qualifications, training gaps, time exceptions, "do we need a third crew?" scenarios.

Boundary: no hiring, firing or pay decisions — ever
GuardAI roadmap

Governance intelligence: privacy posture, contract & claim review, regulatory changes — packaged for human professionals.

Boundary: prepares review packets · professional sign-off stays human
Opportunity OS roadmap

Finds what the business could do next — new services, territories, pricing — with staged evidence, fit checks and supervised experiments.

Boundary: detects, never executes · cheap evidence before paid research

Why this architecture is fundable

Incumbents lock AI behind $500+/month tiers; point tools cost $25–$300/month each. Service OS ships the governed core at one flat price and expands per-tenant revenue through optional packs — same data fabric, near-zero marginal build cost per pack, and a category-pack compiler that turns one platform into HVAC, cleaning, roofing and 17 more verticals without new codebases.

Business eventsGoverned read toolsBriefing & answersOwner decidesOutcome measured

◈ NexusOS — daily briefing

✦ Tuesday 7:00 AM
Money: $11,860 collected MTD · 5.2 weeks payroll runway [AuditMax]
Attention: 1 new lead · 1 overdue invoice · 1 credential expiring [exceptions]
Opportunity: 9 jobs declined for capacity (~$2,400) [leads]
Decision pending: third-crew scenario awaits your review. No actions taken.

Scenario: add a third crew?

Break-even: ~14 jobs/mo · est. cost $4,100/mo
Demand evidence: 91% crew utilization + declined jobs. Unknown flagged: local wage rates. Scenario is isolated — touches no live data.
awaiting owner
Job completedFeedback ask (every customer)Reviews & complaints inRecovery case + draftOwner posts/remedies

◈ ReviewArmor — reputation inbox

Google · ★★★★★ · "Crew was amazing"
Matched to job #1042 · reply drafted, one tap to post
draft ready
Google · ★★ · "missed the baseboards"
Recovery case opened · re-clean offer drafted · private data stripped from public reply
A5 · your call

Honest by design

Every customer gets the same ask
No review-gating, no cherry-picking. Compensation and public replies are separate human actions — never automatic.
Invoice / payment / expense eventsAppend-only ledgerAR · cash · profitabilityReconciliationFinance exceptions to owner

◈ AuditMax — money health

Cash runway
5.2 wks
Receivables
$1,240
Avg job margin
38%
Payout mismatch: Stripe payout $1,612 vs expected $1,637
Candidate cause: dispute fee on #1038 · evidence attached · one-click confirm
exception

Deterministic by rule

AI explains, math computes
Every figure reproducible from append-only events in integer cents. No money movement, no AI-invented amounts — corrections are linked reversals, never edits.
Validated address + service codesProvider-verified rateEvidence stored & cachedGaps & exceptionsPro-review packet

◈ Tax Intelligence — readiness

June liability: $0 collected (SF residential cleaning — non-taxable)
412 calculations · 89% served from cache (same address + codes) · evidence on every line
!
Address outside registered states: Reno, NV inquiry
Finalization blocked — never silently zero-taxed. Flagged for your registration decision.
blocked

Hard limits

No filing, no payment, no tax positions
ZIP codes never finalize a rate. The pack prepares a CPA-ready packet; a human professional signs — the AI cannot.
Time · credentials · availabilityCapacity & gapsTraining / hiring scenariosOwner decides — always

◈ TalentForce — capacity

Crew utilization
91%
Declined jobs (4 wks)
9
Credentials expiring
1
RT
Rosa — cleaning cert expires in 12 days
Auto-drops from eligible lists on expiry · renewal link prepared
action soon

A6 — never autonomous

No ranking, hiring, firing, pay or discipline by AI
Indicators carry context and cannot trigger employment action. Scenarios inform the owner; people decisions stay human by architecture.
Policies & regulationsDeterministic checksRisk findings + citationsChange proposalProfessional sign-off (human)

◈ GuardAI — governance watch

§
SF paid-sick-leave notice updated
Affects worker docs · proposal drafted with source citation · counsel packet ready
review
Customer-consent audit: 100% of SMS contacts have recorded basis
TCPA posture green · quiet hours enforced platform-wide

Assist, never authorize

AI prepares; professionals approve
Review packets carry exact versions, scope and dates. Sign-off identity is human, recorded, and supersedable — AI cannot create professional authority.
Internal signalFree/official evidencePaid research (approved $)Fit & readinessOwner decisionSupervised experiment → outcome

◈ Opportunity OS — opportunity card

Add move-out cleaning as a service line
Evidence: 14 requests declined in 90 days · est. range $380–$520/job · crews already qualified · supplies in stock
readiness: high
demand ●●●○fit ●●●●risk ●○○○freshness 6d
Proposed experiment30-day pilot · 10-job cap · $0 ad spend · stop condition: margin < 30%. No external action was taken.

Evidence before spend

Internal → cached → official/free → approved paid
No opaque score, no 24/7 scraping. Stale evidence decays; contradictions lower confidence; detection can never jump to execution.
Research questionOfficial & licensed sourcesObservations + provenanceReport with unknownsOwner acts

◈ GeoPulse — territory brief: Daly City

Households in service radius: 31,400 [US Census · 2024]
Competing cleaners listed: 12 · median home age 47 yrs · winter demand dip −18% [official sources · cached 9d]
?
Unknown: local wage benchmark
Licensed source available — est. cost $0.18 · requires your approval to spend
approve $

Provenance or it doesn't exist

Every observation: source, date, licence, cost, freshness
External estimates never overwrite verified tenant facts. Reports state contradictions and end with "no action taken."
First-party outcomesAttribution & funnelCustomer voiceGrowth scenarioOwner launches — AI never spends

◈ VibeLocal — where customers come from

Google Business
41%
books at 2.3×
Referrals
28%
best margin
Unknown source
9%
shown, not hidden
Customer voice: "eco products" mentioned 11× this month
Scenario drafted: green-cleaning add-on at $15 — projected from real acceptance rates, ranges not promises
your call

Growth without gambling

No automatic ads, bids, budgets or publishing
Built on outcomes the tenant owns — lead → job → payment — not vanity metrics. Content drafts appear only when asked.

Set up a business express path · category defaults do the heavy lifting

1 · Category
2 · Defaults
3 · Services
4 · Activate

What kind of business is this? One platform, 20 category families — each compiles into its own terminology, workflows and tools.

Residential Cleaningrecurring routes & teams
HVACdispatch & equipment
Roofingprojects & milestones
Plumbingdispatch & inspections
Landscapingrecurring routes
17 more…legal, freight, childcare — readiness-gated

We pre-configured everything from the Residential Cleaning pack. Change anything, or just keep going.

Your words: "Visits", "Crews", "Service plans"
Terminology compiled from the category pack — no technician/job jargon forced on you
default
Workflow: Recurring Route (W2)
Plan → recurrence → route → visit → checklist → billing → next visit
default
AI: answer missed calls & texts, hold slots, draft quotes
Consequential actions always come back to you · $12/mo AI budget cap
default
Service area: San Francisco + 15 mi
Leads outside the area are politely declined, never guessed at
edit later

Starter price book — typical local ranges included. You can refine anytime; the AI quotes ranges from this book and never invents a price.

Standard clean (weekly / bi-weekly)$110 – $140
Deep clean$280 – $340
Move-in / move-out$380 – $520
Inside oven & fridge add-on$45
Tenant Operating Blueprint · v1 · #7e2a

Ready to activate Brightside Cleaning Co

Validated: dependencies ✓ · workflows reachable ✓ · permissions ✓ · nothing paid or external was turned on. Connect payments and your business number whenever you're ready — the system runs in safe mode until then.

✓ Active — exact version approved by Owner, logged forever

Total setup time in the live product: about 5 minutes. Compare: legacy field-service tools average days of configuration and onboarding calls.

The field side — one adaptive app for every trade offline-first · workers see only their assigned minimum

Rosa · Crew A ● synced
Today — 3 visits
Thu Jun 18 · route optimized · 23 min total drive
11:00 · Deep clean — Sara K.
3bd/2ba · Noe Valley · first visit
🔑 Door code unlocks at 10:45 — hidden until day-of, wiped after the visit
Clock in · 📍 one-time location stamp
Checklist · 12 items
Kitchen — counters & sink
Inside oven & fridge (add-on)
Bathrooms ×2 + photo
Baseboards — all rooms

No live tracking, no cameras — a single disclosed location stamp at clock-in/out. Access codes are assignment-scoped and purge on completion.

Rosa · Crew A ⇡ offline · 3 queued
Visit complete — almost done
2 photos attached · 1 part used: eco supplies kit
Customer sign-off
Sara K.
binds to checklist v3 · time-stamped
Complete visit
No signal? Everything queues in a durable outbox and syncs exactly once when you're back online — nothing is ever lost or duplicated.

Same app becomes a technician's (HVAC) or a crew lead's (roofing) — terminology and screens compile from the category pack.

What the customer experiences no account · no app · no login — just a secure link by text

Brightside Cleaning secure link
Quote Q-118 · $345.00
Deep clean 3bd/2ba + oven & fridge · Thursday 11 AM held for you
"Our two-person crew will deep-clean every room top to bottom — your home guest-ready before Thursday evening."
Approve with your signature
Sara K.
Approval binds to this exact version (#a41c). If anything changes, you'll re-approve.
Approve & book — $345.00

One tap from the text thread. Typed-signature approval is ESIGN-valid and locked to the document version.

Invoice #1044 job complete ✓
Deep clean — $345.00
Completed 1:52 PM · 2 photos · signed checklist attached
Add a tip for Rosa & Tina?
none10%15% · $51.7520%
Pay $396.75 · Stripe secure
Card handled by Stripe — the platform never sees it. Save a card to enable auto-pay for your weekly plan (you control it, with notice before every charge).

Pay link with tip → money lands in the owner's own Stripe account. Auto-pay is a customer mandate, never an AI decision.

⌗ AI & Automation Control Tower Layer 5 — one place to see, budget and stop every automation

AutomationLevelRuns · 30dCostOn
Missed-call text-backA438$0.41
Lead qualification & slot holdA2–A331$1.12
Quote narrative draftingA322$0.66
Review request after jobA454$0.58
Overdue-invoice remindersA3 → A56 drafted$0.09
Public review repliesA52 awaiting you$0.05

Every automation: registered task · exact policy version · per-run cost attribution · deterministic kill switch. Toggling off stops it instantly — the business keeps running without AI.

Spend & routing

AI spend MTD
$3.84 / $12 cap
Model routing by task
Cheap fast model for classification · stronger model for drafting · provider-neutral gateway, swappable without code changes
Hard stops
Daily ceiling · per-task ceiling · anomaly circuit breaker. At any limit, AI pauses and the Exception Center tells you why.
Evaluation gates
Zero tolerance in tests: no invented money, no cross-tenant leak, no fake sign-off. A failing eval blocks release, not the customer.
Service OS — the governed AI back office for small service businesses, at one flat honest price. Interactive prototype with sample data · no backend · built to show the working model: AI proposes → deterministic policy verifies → the owner decides.
✦ Ask your business live · cites records
✦ Assistant
Hi! I'm here on every screen. Ask me anything — cash, schedule, leads, taxes, reviews — and I'll answer from your records. Try: "who owes me money?"