Six pillars · One tool · Every trade
What Goliath OS does
Each section below corresponds to a pillar your shop already runs today — usually as five different tools held together with duct tape. Goliath OS replaces all of them.
Pillar 1
Dispatch + Schedule that actually run on a phone
Drag-and-drop kanban for office. Swipeable mobile views for the field. Live tech GPS, customer ETA tracking page, and an AI that proposes a smart re-shuffle the moment a visit cancels.
Live tech GPS + customer ETA page
Customers get a /track/[token] link with the tech's live position. Same UX as DoorDash. Eliminates 80% of "are they coming?" inbound calls.
4-week capacity forecast
Tech-by-tech heatmap with AI callouts: "Week 2 you're 130% booked on installs — consider OT or shifting jobs." Bottleneck-skill flags when you're short on EPA-cert techs.
Smart auto-rebalance
When a visit cancels or a tech calls out, AI drafts a one-click reshuffle of the day's remaining work. You approve or dismiss — never auto-applies.
Route-optimize per tech
2-opt nearest-neighbor pass on each tech's stops, respecting arrival windows. Saves 15-40 minutes of drive-time per tech per day on average.
Pillar 2
AI Receptionist that books real appointments
Twilio + Claude Sonnet. Picks up after-hours, identifies as AI on first turn (Bill 96 + transparency-by-default), books a real Visit on the calendar, and hands every emergency to a human within seconds.
Books on your real calendar
Not a chatbot stub — calls /api/visits, respects capacity + arrival windows + skill requirements. The booked Visit shows up in dispatch instantly.
Hard guardrail on payments
Never reads card numbers aloud. Sends a Stripe payment link via SMS to the caller. Customer pays with Apple Pay / Google Pay in 4 taps.
Multi-language
en-US / en-CA / fr-CA / es-419. Quebec area-code fallback to fr-CA per Bill 96. Disclosure greeting localized.
Emergency escalation
Gas / fire / flood / CO / electrocution keywords short-circuit the LLM and page the on-call tech via SMS within 2 rings.
Pillar 3
Estimates → Invoices that close themselves
Multi-option (Good / Better / Best) on every quote. Inline price book. Auto-tax per US state or Canadian province. Stripe payment links via SMS. AI-drafted dunning when invoices age.
AI Estimate from symptom
Customer types the problem on the public booking page. AI drafts a 3-tier estimate from your real price book + RAG manuals. You approve before send — never auto-publishes.
Customer e-signs on phone
Drawn-signature canvas. Locked PDF generated server-side. "Approved" stamps stored with IP + UA for audit.
AI dunning when AR ages
Tone escalates Friendly → Firm → Final-Notice → Escalate as days-overdue grows. You approve every send. Bilingual (en + fr-CA).
Auto-tax per jurisdiction
GST/HST/PST per Canadian province. State + local sales tax in the US. Pulled from the customer's service address, never the billing address.
Pillar 4
A mobile app that works in dead zones
Offline-first. IndexedDB-backed write queue. Today's-Brief auto-prefetches every customer / job / address / equipment / pricebook on first online hit. Voice copilot. Photo capture with smart-tag AI.
Today's-Brief
~3-8 MB IndexedDB cache vs ServiceTitan's 100+ MB manual download. Bundles every customer + job + address + history + pricebook on first online hit. Auto-fires.
Smart photo gate
Required-photo checklist per job type. Claude Vision tags + OCRs equipment plates. Replaces VisionCam + a clipboard.
Voice copilot
Tech says "add 30 minutes to the bill, customer signed off". Copilot adds the line, parses the price book, drafts the customer summary, requests signature.
Offline write queue
Photos / forms / time entries / status updates queue in IndexedDB. Drains in FIFO order on reconnect with idempotency keys — zero data loss.
Pillar 5
Symptom-to-Cause RAG (real RAG, not autocomplete)
Tech says "Carrier furnace, no heat, error code 33". AI retrieves the actual page from the indexed manual + your private equipment specs, gives a 90%-confidence top-cause + parts list, with [#1] [#2] citations back to the source page.
Real pgvector RAG
Native Postgres vector index in Neon. ivfflat cosine search. Hybrid: indexed manufacturer manuals + your private upload corpus. No external Pinecone bill.
[#N] inline citations
Every claim links to the manual page it came from. "Open original PDF" for tech to verify before they replace a $1,200 board.
Parts pre-filled
AI maps the diagnosis to your real price book + on-hand inventory. "You probably need a flame sensor — you have 3 on truck #7."
Escalate to office
Low-confidence diagnosis gets an explicit RED chip and a one-tap "Ask office" that posts the question into the #techs channel with the photo + reading attached.
Pillar 6
A full accounting spine — not a QBO bridge
Native general ledger. Stripe + Plaid bank feeds. Three-way match. AIA G702 / G703 progress billing for commercial. Lien waivers. Auto-categorize transactions. AI-flagged journal-entry review. Replace your QBO without the migration pain.
Real GL spine
23-step accounting spine, JournalSourceType for every entry, void/repost, full audit trail. The kind of accounting your CPA actually wants to look at.
Bank + CC reconciliation
Plaid for US banks, Flinks for Canadian. CSV upload fallback. AI auto-suggests the GL account based on your VendorMemoryMap. ≥0.85 confidence + 3 prior matches → auto-applies.
Three-way match
PO ↔ Receipt ↔ Bill anomaly detector. Flags overcharges / quantity mismatches / missing receipts / duplicate bills before you pay them.
QBO migration in 4 steps
Accounts CSV → customers CSV → vendors CSV → trial balance cutover. Reversible at every step. Most shops are migrated inside 90 minutes.
See it on your shop's data.
14-day free trial. Migration done in an hour. We'll do a live walkthrough on your stack if you ask.