CrosstownOS™
Crosstown Acquisitions
Every property management system in this vertical was built to manage reservations. CrosstownOS was built to operate assets — from the first off-market lead to the last delinquency notice.
The Silver Tsunami is not a metaphor — it is an active transfer of title.
Manufactured Housing Communities (MHCs) positioned across the United States. The overwhelming majority are controlled by first-generation owner-operators approaching retirement with no succession infrastructure, no institutional buyer pipeline, and no digital operating system in place.
Annual outdoor hospitality market footprint. Over 52 million North American households participate in campground and RV recreation annually. Infrastructure development is years behind demand, and incumbent property management platforms cannot support the yield optimization required by institutional capital.
Active campground households driving the demand side. The supply side — park inventory — is aging, fragmented, and undercapitalized. The operators who own it are sellers. The institutions who want it need an operating system before they can operate it.
Generational Transfer Has Begun
The transfer of MHC ownership from founder-operators to institutional capital is not a prediction — it is an active process measurable at the county recorder level across every major U.S. market.
Structural Entry Friction Is Unsolved
Title, condition, financial normalization, and operational infrastructure are all manual, disconnected, and executed differently by every buyer and broker in the market.
Incumbents Are Not Solving This
Campspot, ResNexus, and RMS Cloud are reservation-calendar products. None were built for institutional acquisition, asset-level financial intelligence, or the requirements of a GP running a leveraged portfolio.
The incumbents own the booking calendar. They don't own the asset.
Incumbent architecture
- Reservation-First Models:Campspot, ResNexus, Newbook, and RMS Cloud are reservation-first systems. Their data models track availability slots and transactional revenue — not asset condition, lease compliance, delinquency aging, or infrastructure lifecycle.
- Underwriting Deficit:No underwriting engine, no deal pipeline, no document ingestion, no capital structure modeling, no LP-ready reporting. They cannot answer: “What is this property worth at acquisition, and what does it cost to stabilize it?”
- Gated Ecosystems:API access is gated, expensive, and architected for bolt-on integrations — not event-driven cross-system intelligence. No webhooks. No pub/sub. No multi-context agent execution.
- Siloed Portfolios:Operator data is siloed inside each PMS instance. No portfolio-level view, no cross-property delinquency reporting, no materialized intelligence surface for institutional capital.
Operator reality
- Fragmented Tooling:A park owner managing 150 spaces runs 4–6 disconnected systems concurrently: a PMS for reservations, QuickBooks for ledgers, phone and email for vendors, a spreadsheet for underwriting, and word-of-mouth for capital.
- Field Isolation:Field operators in rural environments have no mobile-native tools, no offline persistence, no voice-to-action capability, no geofenced vendor tracking. A maintenance issue requires a call, a text, another call, and a mental note.
- Forensic Due Diligence:Institutional buyers conducting due diligence on a 100-space MHP currently receive a 3-tab Excel file, a hand-annotated rent roll, and a T12 that does not reconcile with the bank statements.
- Zero Standardization:There is no standard data normalization layer in this vertical. Every acquisition is a manual forensic audit. Every transition is a platform migration.
The structural gap
- Ununified Stack:No single platform connects deal origination → underwriting → operations → capital exit in one event-driven architecture. Every operator runs a fragmented stack. Every institution builds the stack themselves, per deal.
- Sovereign Opportunity:The market is ready for a sovereign operating layer. The buyers are capitalized. The sellers are motivated. The assets are being transferred. The platform that owns the data layer owns the market.
“CrosstownOS is not the best PMS for outdoor hospitality. It is the first operating system for the outdoor hospitality asset class.”
The gap between how these assets are managed today and how they need to be managed at institutional scale is CrosstownOS's entire addressable market.
CrosstownOS is not a better PMS. It replaces the PMS architecture itself.
The incumbents built monolithic systems. Every user workflow bends to fit the PMS data model. CrosstownOS inverts this. The asset operates the system — not the other way around.
Event-driven core
Every domain change — a lease status update, a utility meter anomaly, a deal stage transition, a work-order SLA breach — emits a canonical event to the domain_events table. External systems integrate via stable event schema, not point-to-point APIs. No polling. No stale data. No manual sync.
Agentic operations
23 AI agents subscribe to the event bus. Agents triage leads, underwrite deals, monitor SLA compliance, detect utility anomalies, dispatch vendors, and orchestrate resident communications — without human initiation on routine execution. Not rule-engine automation: real-time, context-loaded intelligence routing.
Unified space & device model
Every physical element — RV pads, MHP lots, utility meters, gates — is a first-class entity in a living knowledge graph. Not a field in a reservation system. A sovereign data node with its own state machine, event history, and telemetry binding.
Isolated domain workspaces
Sourcing, Deal Desk, Park Operations, Dispositions, and Comms Hub are domain-isolated workspaces. Each owns its aggregate roots. All cross-context communication flows exclusively through domain_events. No tangled schemas. No cross-workspace direct table queries. No monolith.
Portfolio intelligence surface
Real-time materialized views for institutional capital: NOI by property, collections velocity, occupancy stabilization rate, and LP-ready reporting with NDA-gated data room access. PE buyers do not reconcile CSV files. They log in and see current portfolio reality.
CrosstownOS architectures around events, agents, and asset-level consciousness. The park operates itself. Capital optimizes itself. Humans make decisions while software executes the consequences.
The Sovereign Stack — every layer earns its place or it does not ship.
Next.js 15 App Router — Vercel Edge Runtime
Role-based workspace dashboards, agent command palette, real-time operational telemetry, mobile PWA field operations layer.
Trigger.dev v4
Event routing, agent task dispatch, document ingestion, portfolio analytics aggregation, all scheduled background work — n8n fully deprecated and migrated.
Supabase / PostgreSQL — Raw Parameterized SQL
Multi-tenant SSOT; numeric financial precision; zero ORM; zero abstraction layer between application logic and data.
NextAuth.js — Credentials + MFA
RBAC roles: internal_admin, lp_partner, whitelabel_operator, camp_host; MFA enforced for internal_admin and lp_partner.
Supabase Realtime + Yjs
Live operational dashboards, collaborative deal editing with active-user awareness, field sync state broadcasting.
Multi-provider LLM routing via 9router
23-agent dispatch; OpenAI + Anthropic routing; NEPQ / Tactical Empathy / Challenger Sale frameworks injected at prompt assembly runtime.
Iron Vault Integrity
All database access via raw parameterized SQL using the native pg driver with explicit positional parameters ($1, $2) and prepare: false. No ORM. No query builder. No abstraction layer. Zod schema validation enforced at the application boundary only.
Trigger.dev as Conductor
All workflow automation, agent dispatch, background scheduling, and notification logic operates exclusively through Trigger.dev v4. No inline notification logic in API routes. No n8n workflows in production. Trigger.dev is the singular Conductor of all async execution.
Multi-Tenant Isolation Guarantee
org_id is present on every table and injected into every SQL query. Row-Level Security policies enforce isolation at the database layer. Cross-tenant data leakage is structurally impossible — not policy-dependent. Every white-label tenant is fully isolated.
Engineering Governance · Raw SQL · Event-Driven · Multi-Tenant · Zero ORM
“The Deal” is the platform. Everything else orbits it.
The pipeline_deals table is the Central Core Aggregate Root of CrosstownOS. It is not a CRM record. It is the 141-field Institutional Underwriting Matrix — the single system of record for every acquisition analysis, document event, and capital decision on the platform.
Deal Identification
Legacy Financial Performance
Crosstown Pro Forma
Operating Expenses, Detailed
Infrastructure & Utilities
Capital Structure
Strategic Flags · Docs · Profile · Pipeline
deal_name, asset_name, location, market, park_type, total_acreage, pad_count_total, poh_count, deal_source, deal_stage
gross_revenue, noi_legacy, t12_noi, occupancy, asking_price, gross_potential_lot_rent, rubs_utility_reimbursement, effective_gross_income, +7 income stream variables
noi_crosstown, shadow_value, terminal_value, levered_irr, entry_cap_rate, exit_cap_rate, coc_return, stabilization_progress (computed), vacant_pad_fill_pace_per_month, annual_rent_growth_percentage, +10 forward-model variables
payroll_maintenance, payroll_admin_benefits, utility_trash, utility_electric, utility_water_sewer, maintenance_roads, ongoing_capital_reserves, legal_professional_fees, property_insurance_annual, +16 line-item expense fields
water_metering_type, sewer_infrastructure_type, pedestal_amperage_topology, pipe_material_composition, road_surface_material, lift_station_count, infrastructure_score (computed), +8 physical condition variables
loan_to_value_ltv, gp_equity_split, lp_equity_split, preferred_return_percentage, gp_promote_tier_percentage, debt_service_coverage, interest_only_period_months, target_refi_year, +12 capital architecture fields
System Invariants: Every field mutation increments a version counter. Every stage transition writes synchronously to domain_events before async dispatch. All financial fields use numeric precision — never integer arithmetic. Every document ingestion triggers an async underwrite-deal.ts recalculation. The Deal never silently goes stale.
Sourcing: off-market assets don't advertise. We find them first.
Public Records + County CADs
market-research.ts
Agent 05 Scraper
maxTries: 4 · exponential backoff
35-Point Scoring Engine
indexed · sub-500ms queries
Triage Hopper
gate: 22/35 threshold
1-Click Promotion Bridge
sql.begin() / sql.commit()
pipeline_deals
sourcing.lead_qualified
Autonomous multi-agent lead scraper
Agent 05 (Data Analyst) executes crash-resilient, rate-limit-shielded extraction across public records, county CAD systems, and U.S. Census datasets. Runs maxTries: 4 with exponential backoff and regex-based parsing for unstructured public records. Every discovered asset writes to market_research with full domain data: owner contact, estimated pad count, parcel ID, title keywords, coordinates, and review metadata. Multi-source aggregation with automatic deduplication; partial failures do not corrupt the record set.
35-point automated market scoring matrix
Every lead is scored against 35 weighted dimensions: occupancy signals, star ratings and review volume, title nomenclature analysis, and asset class. Negative weight applies when title fields contain institutional signals (LLC, HOLDINGS, VENTURES). Configurable per asset class — MHP, RV, or Hybrid. Stored as an indexed column for sub-500ms filtered queries at scale. Leads scoring ≥22/35 qualify for the Triage Hopper.
Triage hopper viewport — inbox-zero acquisition intelligence
A columnar data grid displaying only records at or above the 22/35 threshold. Bulk-select, inline edit, sortable headers, and a seasonal data hygiene matrix enforce discipline at scale. Records move through Pending → Reviewed → Deferred → Promoted. Mobile Swipe-to-Promote (FIELD-003) lets field operators triage via haptic swipe cards — swipe right triggers atomic promotion; swipe left archives with a configurable re-surface schedule.
1-click bare-metal promotion bridge
A single atomic SQL transaction — sql.begin() / sql.commit() — migrates the market_research record into pipeline_deals with full schema-mapped population, auto-tagging to prevent re-ingestion, deduplication by parcel ID and APN, and immediate sourcing.lead_qualified emission to domain_events. Zero partial promotions under any failure scenario. Competitive proximity context is appended automatically on promotion (DD-011).
Every T12 that lands on this platform is parsed, verified, and underwritten — without human intervention.
pending
enqueued on upload
processing
locked_at / locked_by
completed
schema-mapped
failed → retry ×4
max_attempts: 4 · backoff
large-1x · 4 vCPU / 8 GB · 10-min window · concurrency 3
Automated T12 & document ingestion queue
Uploaded documents — broker PDFs of 50+ pages, multi-tab rent rolls, scanned P&Ls, unformatted Excel — enter document_ingestion_queue and enqueue immediately. The queue enforces a full state machine: pending → processing → completed/failed. Retry logic with attempt_count tracking and exponential backoff up to 4 attempts. A locked_at / locked_by mechanism prevents duplicate processing across parallel workers. Machine sizing: large-1x (4 vCPU, 8 GB RAM), 10-minute max execution window. Queue concurrency: 3 parallel workers.
Vision AI parsing pipeline
The Document Ingestion Processor (07_DOC) runs a Vision AI extraction pipeline against PDFs and unformatted Excel. The parser maps extracted financials to Crosstown’s standardized Chart of Accounts and the 141-field schema — auto-populating poh_rent_income, utility_trash, payroll_maintenance, vacancy_credit_loss_percentage, other_ancillary_income, maintenance_roads, and all applicable expense lines. Broker-stated metrics are cross-referenced against extracted actuals; material discrepancies trigger a deal.compliance_flagged event and a structured discrepancy report on the deal card.
Real-time underwriting recalculation engine
The Live Underwriter (07_LIVE) subscribes to deal.document_ingested, utility.reading_anomalous, and lease.status_changed. On any trigger, underwrite-deal.ts executes — recalculating DSCR, MAO (sensitivity grid across vacancy ±5%, rate ±100bps, expense ±10%), Levered IRR (bisection algorithm, convergence within 50 iterations at <0.01% tolerance), Cash-on-Cash, Cap Rate, Equity Multiple, and the preferred return waterfall. Machine sizing: medium-2x (2 vCPU, 4 GB RAM), 5-minute window. Every recalculation writes a versioned snapshot.
AI document comparison & generation suite
Counterparty redlines are processed through the AI Document Comparison Engine (DD-009) — a side-by-side diff with an AI-generated plain-English summary of every material change, rendered in <CrosstownEditor /> for mandatory human review. The OM & Contract Generation suite (DD-008) produces LOIs, Offering Memorandums, 3-Day Notices, Lease Agreements, and Purchase Agreements from live pipeline_deals data via Agents 09, 16, and 17 — one-click PDF export on branded letterhead. No document exits without clearing the human review gate.
Dispositions: deal-ready capital, matched by algorithm — not rolodex.
Capital partner matching engine
On dispositions stage, a deal.stage_changed event triggers dispositions-capital-match.ts. The engine scores every record in deal_partner_matches and contacts_partners — capital availability, target market, asset class strategy, hold period alignment, and historical co-invest — against the live underwriting profile. A fit_score is computed per partner and surfaced via the Dynamic Capital Roster (DynamicCapitalRoster.tsx). From Phase 1, operational telemetry — utility recapture, stabilization velocity, collection performance — feeds match scoring, a data advantage cold capital introduction cannot replicate.
Syndication waterfall modeler
SyndicationWaterfall.tsx models full LP/GP distribution structures directly from capital structure fields: lp_equity_split, gp_equity_split, preferred_return_percentage, gp_promote_tier_percentage, sponsor_total_comp, and debt_service_coverage. All calculations use numeric precision — never integer arithmetic. Distribution tiers, preferred return hurdles, and promote triggers are modeled visually. The Polymorphic Matrix (PolymorphicMatrix.tsx) enables multi-dimensional capital structure comparison and stress-testing from identical underwriting inputs.
Joint venture & wholesale execution layer
Concurrent exit and deal-share structures are handled through the Joint Venture Modeler (JointVentureModeler.tsx) and Wholesale Spread Calculator (WholesaleSpreadCalculator.tsx). Every JV scenario, LP raise, and wholesale assignment models against the same pipeline_deals core data — not a spreadsheet that drifts. Agent 17 (Dispositions) operates the desk, with Agent 09 (Deal Structurer) on transaction architecture and Agent 10 (Partnership Director) managing capital relationship outreach.
LP / partner data room
A gated repository accessible exclusively to the lp_partner role. NDA digital acknowledgment is logged to audit_event_log before any access is granted — the record is immutable by design. The room houses deal decks, signed agreements, K-1s, and capital account statements. Download activity is tracked per LP, per document, per version. LP notification fires within 5 minutes of any new internal_admin upload. The full interaction history constitutes a clean, non-deletable due diligence paper trail.
100% LP until capital returned
pref hurdle — preferred_return_percentage
gp_promote_tier_percentage
lp_equity_split / gp_equity_split
Deal-ready capital, matched by algorithm — not rolodex. The same underwriting core powers every JV, LP raise, and wholesale assignment.
Unified property control across one space registry.
property_sites
site_id, org_id, type, status, coordinates
leases
lease_id, site_id, lease_type, rent, state
utility_meters
meter_id, site_id, meter_type, serial
utility_readings
reading_id, meter_id, value, ts
The Space Registry Core
Real-time data persistence layer for tracking physical infrastructure, mapping properties across a standardized entity schema: property_sites, leases, utility_meters, and utility_readings.
Unified Space Model Boundaries
Complete physical asset normalization across structural space variables: rv_pad, mhp_lot, cabin, and amenity.
State-Machine Enforcement
Automated state transitions governing the physical asset lifecycle, eliminating un-monitored vacancies by enforcing rigid state boundaries: available, occupied, maintenance, offline, and seasonal_hold.
Strict Lease Binding
Automated lot-rent and occupancy tracking via relational lease constraints: transient_rv, seasonal, annual_mhp, and short_term_rental.
| Site | Type | Lease | State | Rent |
|---|---|---|---|---|
| A-014 | mhp_lot | annual_mhp | occupied | $525 |
| A-021 | rv_pad | transient_rv | available | — |
| B-007 | rv_pad | seasonal | seasonal_hold | $410 |
| B-033 | cabin | short_term_rental | maintenance | $0 |
| C-002 | mhp_lot | annual_mhp | occupied | $540 |
| C-018 | amenity | — | offline | — |
Delinquency & ledger control, posted automatically.
| Date | Description | Debit | Credit | Balance |
|---|---|---|---|---|
| 06-01 | Lot rent — A-014 | $525.00 | $525.00 | |
| 06-01 | Payment received | $525.00 | $0.00 | |
| 06-01 | Lot rent — C-002 | $540.00 | $540.00 | |
| 06-06 | Late fee auto-post | $50.00 | $590.00 | |
| 06-10 | Partial payment | $300.00 | $290.00 |
Double-Entry Financial Architecture
Real-time financial transaction tracking, logging every debit and credit ledger change across rent charges, adjustments, and collections.
Automated Late-Fee Engine
Nightly background processes executed at 1:00 AM local property time via scheduled tasks, checking lease rules and posting fees with zero manager intervention.
Aged Delinquency Reporting
Programmatic data compilation that partitions outstanding balances into structured aging buckets: Current, 30 Days, 60 Days, and 90+ Days Past Due.
Legal Action Readiness
One-click data exportation providing clean, court-admissible CSV transaction ledgers for immediate eviction attorney handoffs.
Live-yield digital twin & telemetry.
Hover a lot node to inspect live condition · filter by status above.
Active lot rent performing at or above regional pro forma market-rate thresholds.
Active below-market tenant lease revealing immediate rent-stabilization upside.
Active lease state flags severe collection loss or past-due lease balances.
Available lot asset ready for immediate occupancy or home placement.
Active maintenance ticket or physical utility constraint tracking.
Live Geospatial Node Mapping
Dynamic, property-level visual dashboard rendering physical plot structures as live interactive data nodes using exact coordinate data.
Direct Hardware Telemetry Ingestion
Direct data capture from on-site digital hardware layers, monitoring water flow rates at 15-minute intervals to catch resource leaks.
Autonomous Anomaly Defense
Twenty-four-hour background scanners analyzing usage data; detecting abnormal consumption (e.g., 3:00 AM water surges) and routing incident context directly to on-call teams via Telegram alerts.
Built for one bar of signal and zero margin for error.
Disconnected environment persistence
Mobile-native progressive web application (PWA) built specifically for low-connectivity rural environments, ensuring system performance on 1 bar of cellular signal.
Local state write-ahead logging
Enforcing data transaction persistence by recording all field operator data entries sequentially inside a local IndexedDB mutation queue before hitting network layers.
Zero-loss ingestion pipeline
Automated queue synchronization protocols that detect cellular recovery events and securely flush local mutations to the database via secure endpoints.
Optimistic concurrency resolution
Resolving data conflicts from concurrent offline edits via deterministic Last-Write-Wins (LWW) parameters while generating non-deletable records inside the append-only log.
Work order — B-033
Meter read — A-014
Lease note — C-002
Photo upload — gate
Write-ahead log · LWW conflict resolution · Append-only mutation architecture
The voice-to-action engine. Speak the field into the system.
Single-handed thumb-zone focus
UI design optimized for rural field operations — placing all primary navigation, inputs, and controls inside the single-handed lower screen region.
Sub-second audio processing
Native microphone control leveraging edge transcription services to translate natural voice commands into structured data arrays in under 500ms.
Hardware-accelerated ingestion
One-tap native camera execution capturing unstructured T12 financial records, rent rolls, or field asset signs for automated optical character recognition (OCR) parsing.
Acquisitions
Live walk-and-talk deal matrix adjustments updating pro forma models instantly.
Operations
Immediate field dispatch commands, maintenance updates, and vendor assignments.
SOP Generation
Verbatim on-site process recording converted directly into structured procedural manuals.
Awaiting voice command
Tap and speak. Output routes in under 500ms.
Emergency dispatch & geofencing. P1 bypasses the queue.
Slide to confirm emergency dispatch
Disarmed
Agent 11
Priority contractor dispatch
Agent 13
Geofenced resident broadcast
P1 life-safety panic controller
Slide-to-confirm emergency framework engineered with custom touch resistance to eliminate accidental triggers in the field.
Bypassing corporate gating
Instantaneous dual-agent emergency execution that eliminates manual review delays to protect property and human safety.
Autonomous emergency broadcast
Simultaneous execution routing: Agent 11 dispatches priority emergency contractors via automated notifications while Agent 13 broadcasts life-safety directives to all geofenced resident lines.
Geofenced activity verification
Automated contractor tracking leveraging browser-native geolocation checks on arrival, validating vendor location against property bounds with zero app download requirements.
Grand Central Inbox — one platform. Every conversation. Zero dropped threads.
TENANT_OPS
Maintenance · payments · lease
SELLER_LEAD
Acquisition → Agent 01/09
VENDOR_DISPATCH
Work order → Agent 11
ESCALATION
Hot signal → operator handoff
Omnichannel grand central inbox (Chatwoot)
Every inbound channel — Twilio SMS, Resend email, in-app resident messages, and future OTA aggregation — converges into a single Chatwoot-powered workspace. Inbound messages create or append to conversation_threads within 5 seconds. Thread context binds to at least one entity: property_id, site_id, lease_id, contact_id, or deal_id. The same inbox that managed the seller negotiation pre-close continues to manage the resident relationship post-acquisition — a domain_events context transition from deal_id to property_id on close preserves all history with zero migration.
Conductor intent routing matrix
intent-classifier.ts fires on every inbound message, routing across four isolated pipelines within 3 seconds: TENANT_OPS, SELLER_LEAD, VENDOR_DISPATCH, and ESCALATION. Classification confidence is scored at the task level; messages below threshold escalate to a human operator with the last five messages of context auto-assembled. Routing accuracy target: >85%.
Autonomous sentiment intelligence
Every message receives a sentiment_score on a continuous -1.0 to 1.0 scale. Agent 13 (Community Manager) monitors the unresolved intent sweep — agent_response_id IS NULL — and autonomously handles routine inquiries via Tactical Empathy. Hot-signal detection activates on price discussions, timeline urgency, or contract terms — bypassing autonomous response and triggering an immediate operator alert with a timestamped handoff written to pipeline_deals. SLA: operator response within 15 minutes.
Harvest-to-dispatch closed loop
The full cycle — find, negotiate, manage — operates entirely within the Comms Hub. Agent 05 deposits scored leads (≥22/35). Agent 11 generates hyper-personalized outreach from live market_research data injected with NEPQ and Tactical Empathy. Replies capture in the Grand Central Inbox. Agent 13 triages. Qualified seller conversations escalate to Agent 09. The thread that acquired the property becomes the operational inbox for the first resident maintenance request — no platform change, no migration, no dropped context.
The CMO Terminal — acquisitions-focused brand operations, not a social scheduler.
Q3 acquisition reel
ScheduledOperator testimonial
In ReviewFounding cohort one-pager
ApprovedBrand guidelines v2
DraftDrone footage — Quinlan
PublishedPress kit refresh
DraftSeller nurture sequence
ScheduledLP update — June
In ReviewCMO content terminal architecture
The Content View is the brand operations command center — driven by CMOTerminal.tsx, ContentHopper.tsx, ContentInspectionDesk.tsx, and EditorialCanvas.tsx. All content lives natively in PostgreSQL via content_calendar. No third-party CMS, no external API latency, no vendor lock-in. The aggregate emits canonical events at every stage: content.drafted, content.scheduled, content.published, content.performance_ingested — making publishing a first-class citizen of the event-driven architecture.
Content matrix kanban & state machine
Every asset is managed across four channel columns: Social, Brand, Media, and Outreach. Assets move through Draft → In Review → Approved → Scheduled → Published. Drag-and-drop governs stage transitions with publish-date scheduling at the Kanban level. Phase 1 ships approval workflow automation and auto-publish at scheduled timestamp — zero manual publish actions at go-live.
<CrosstownEditor /> — the universal authoring surface
All platform content — marketing copy, SOP documents, generated LOIs, park website sections, and Comms Hub reply windows — is authored through <CrosstownEditor />, a TipTap-based WYSIWYG editor with bubble menus, AI rewrite, slash commands, image insert, callout blocks, and table builder. Output stores as clean HTML or TipTap JSON in Postgres. Real-time collaborative editing via Yjs + Supabase Realtime. Agent 03 (CMO) operates this workspace.
Marketing asset infrastructure & ISR park websites
Park photography, video, drone footage, and ad creatives are stored under property-scoped paths on Vercel Blob and AWS S3. SEO-optimized headless park websites are generated via Next.js ISR from properties data — no manual page builds. ISR cache invalidates within 60 seconds of any Postgres write affecting a public-facing site. Operators see published changes live within one minute of save. Google Mobile-Friendly compliance enforced at generation.
Multi-party financial architecture — operator economics, executed at transaction speed.
Tenant payment
ACH · card · debit
Stripe Connect split
Configurable per PM agreement
Operator connected account
Net rent distribution
Crosstown master account
Management + SaaS fee capture
- 01Onboarding Form Ingestion
- 02Automated DNS API Propagation
- 03SSL Certificate Generation
- 04Postgres Organization Record Isolation
- 05org_agent_roster Provisioning (23 slots)
- 06Subdomain Live Deployment (< 5 min)
Stripe Connect capital waterfall integration
Every tenant payment — ACH, credit, or debit — executes an automated fund split at transaction time via Stripe Connect. The ratio is configurable per PM agreement: operator connected account receives net rent; Crosstown master account receives the management and SaaS fee. A webhook fires on every successful transaction, triggering a Trigger.dev task that writes a double-entry ledger record: debit_account, credit_account, amount, transaction_id, property_id, timestamp. 100% of successful transactions create ledger entries within 5 seconds. Zero manual reconciliation. Zero AR lag.
Automated corporate ledger syncing
Agent 06 (Controller) categorizes every financial event across the standardized MHP/RV expense fields: bank_fees, software_accounting, payroll_maintenance, utility_trash, marketing_advertising, and all Chart of Accounts categories. Work order costs allocate to property P&L. CapEx is tracked against reserve budgets from ongoing_capital_reserves and capex_upfront_improvements. Daily reconciliation enforces balance accuracy — total debits equal total credits per property per month — verified by ledger-reconciliation.ts.
Multi-property consolidated P&L dashboard
Portfolio income statements cascade from portfolio → property → individual pad with real-time drilldown. RBAC scoping enforces LP visibility — lp_partner accesses only properties within their scope. Metrics include gross rents, vacancy loss, operating expenses, NOI, and net cash flow with MoM and YoY variance. PDF export for LP distribution. Automated monthly P&L generation via scheduled task. Portfolio P&L loads for a 20-property portfolio in under 3 seconds.
1-click white-label subdomain provisioning
A single onboarding script provisions a fully isolated white-label tenant in under 5 minutes: branded subdomain (tenants.parksite.com), SSL certificate, Postgres org record, full org_agent_roster with 23 configurable agent name slots, Stripe Connect split ratio, logo upload, and DNS propagation monitoring with live status. The entire sequence runs on the root Next.js codebase — zero new codebase per tenant. Rate limits configured per tenant in tenant_quota_ledger with soft-limit warning at 80% and graceful degradation at ceiling. Isolation enforced at query level — org_id on every record, every query.
23 AI agents — three execution tiers — one Conductor.
CrosstownOS is not a chatbot with a business card. It is a 23-agent operational intelligence network — each agent defined by a canonical role, a knowledge base in agent_knowledge_base, a Trigger.dev task prefix, and a scoped workspace. Agent 19 does not exist. The numbering gap is intentional and preserved.
Lead Intelligence Classifier
Fires on every unclassified inbound conversation_threads record.
Live Underwriter
Recalculates full pipeline_deals financials on document, utility, and lease events.
SLA Compliance Sweeper
Scans work_orders every 15 min — Emergency 4hr · Urgent 24hr · Routine 72hr.
Utility Anomaly Detector
Hourly statistical analysis against utility_readings; flags abnormal consumption.
Late Fee Poster
Daily at 1:00 AM local; posts late fees per lease terms, zero intervention.
Compliance Policy Enforcer
Evaluates domain_events against compliance_policies; generates approval tasks.
Every agent dispatch, scheduled task, and async execution flows through a single Conductor.
9router Prompt Assembly: at dispatch, (1) pulls KB documents from agent_knowledge_base; (2) injects custom_name from org_agent_roster; (3) applies the assigned Psychological Persuasion Framework — NEPQ, Tactical Empathy, or Challenger Sale. All executions log to audit_event_log with agent ID, payload hash, response hash, latency, and status within 100ms. Human review gate enforced before any external dispatch.
Operator SaaS licensing — the arbitrage is built into the ask.
| Tier | Scope | Rate | Agents |
|---|---|---|---|
| Founding Operator Rate | 1–20 properties · Phase 0–1 cohort | Locked / Grandfathered | 23 included |
| Standard Operator Rate | Phase 1 full-feature parity | Activates Oct 2026 | 23 included |
Phase 0–1 targets independent operators, bootstrapped fundless sponsors, and boutique PE firms managing 1–20 properties. Founding Operator pricing — locked and grandfathered at launch — sits materially below the standard rate that activates at the October 2026 Phase 1 go-to-market. Founding Operator status is finite: the first cohort defines the reference base that funds Phase 2.
Third-party PM companies and boutique investment firms license CrosstownOS under their own brand via the 1-Click White-Label engine (TRE-003) — their own subdomain, their own custom-named 23-agent roster, their own Stripe Connect split, their own isolated org_id scope. Per-tenant monthly revenue stacks on top of direct operator revenue with no additional engineering cost per tenant.
A 75-pad MHP runs 1–2 FTEs covering management, maintenance, bookkeeping, and resident comms. CrosstownOS autonomously executes Agent 11, Agent 06, Agent 13, Agent 08, the SLA Sweeper, the Late Fee Poster, the Anomaly Detector, and Ledger Reconciliation. The economics are not a feature discussion — they are a payroll conversation.
Phase 2 Value Unlock — Telemetry in Capital Matching
When live operational telemetry — utility recapture, collection velocity, stabilization_progress (Group C), and Digital Twin lease health — feeds the deal_partner_matches engine, CrosstownOS becomes the authoritative data source for institutional partners. A compounding data moat: more properties → more precise matching → stronger partner relationships → higher deal velocity.
The sovereign ecosystem tax — when the platform becomes market infrastructure, every transaction is a revenue event.
Phase 3 monetization is not a pricing strategy. It is a structural consequence of operating the only vertically integrated transaction layer between asset owners, operators, capital, and residents in this asset class.
Transaction processing tax
Every financial transaction routed through Stripe Connect — rent, deposits, late fees, vendor invoices, capital distributions — is subject to a 1% ecosystem fee retained at the platform layer. A 50-property portfolio averaging 80 lots at $450/month generates ~$1.8M in monthly volume — yielding ~$18,000/month in platform revenue from a single portfolio, with no incremental feature sold. The fee accrues invisibly at the infrastructure level.
Acquisition sourcing fee
Acquisitions originated within CrosstownOS — scraped by Agent 05, scored by the 35-Point Matrix, promoted via the 1-Click Bridge, underwritten through the 141-field matrix — carry a platform-facilitated sourcing fee at transaction. The provenance chain from market_research → pipeline_deals → domain_events close event creates a clean, auditable attribution record. The deal lifecycle is the audit trail.
NOI value-share
Under full PM agreements, CrosstownOS earns a performance component tied to NOI improvement above the acquisition pro forma baseline captured in noi_legacy and noi_crosstown. shadow_value, stabilization_progress, and terminal_value create the measurement infrastructure. When rents rise above pro forma, delinquency falls below baseline, and utility recapture beats the modeled rubs_utility_reimbursement target — the platform contribution is mathematically attributable and contractually monetizable.
Phase 3 ships only if 50+ properties are operational on Phase 1–2 by month six post-launch. No premature optimization. Revenue and operational density gate further investment.
Jeremy Cross
Managing Partner — Crosstown Acquisitions LLC
Founder — CrosstownOS™
Quinlan, Texas · Hunt County Born
Hunt County roots. Zero outside capital. Every line of code.
Jeremy Cross did not discover the RV park and manufactured housing market from a pitch deck. He grew up in Hunt County, Texas — embedded in these communities from childhood.
CrosstownOS is not a product built from a market thesis. It is a product built from a lifetime of proximity to the asset class, combined with enterprise B2B sales experience — NEPQ, Challenger Sale, and Tactical Empathy trained and deployed commercially across full-cycle AE roles. Every behavioral framework embedded in the agent architecture was first used by Jeremy in a sales call.
- Multi-tenant PostgreSQL architecture with full RBAC and org_id isolation
- 141-field pipeline_deals matrix with Kanban pipeline and version tracking
- Document ingestion queue with state machine, retry logic, and Trigger.dev orchestration
- Capital partner matching with fit_score computation
- 23-agent command palette with 9router prompt assembly
- Collaborative deal editing via TipTap + Yjs
- automation_telemetry monitoring with P1–P4 severity classification
- Full content calendar with ContentZoneView Kanban
- Dispositions Desk with Syndication Waterfall and Joint Venture Modeler
- Complete Trigger.dev Conductor infrastructure
Y Combinator process · Interviews in progress
- Production-grade full-stack engineering — Next.js, PostgreSQL, TypeScript at scale
- Comfort within domain-driven, event-driven, multi-tenant architecture without abstraction shortcuts
- A conviction that physical assets and AI operational infrastructure belong in the same company
- Willingness to build for the operator running properties on one bar of LTE signal with zero margin for error
Institutional seed architecture deploy
- Phase 1 engineering infrastructure — Trigger.dev task suite, PWA Field Ops, Park Operations buildout to October 2026 launch
- Founding operator acquisition, onboarding, and customer success for the first production cohort
- Go-to-market support and institutional partnership development targeting October 2026 entry
CrosstownOS. Dirt and Systems.
Contact: Jeremy Cross · Crosstown Acquisitions LLC · CrosstownOS™
End of Institutional Read-Deck · v1.0 · Confidential & Proprietary