Autopilot Hunter
The end-state workflow: current markets → live properties → ranked call queue → real offer range → buyer transaction evidence. It never labels cached or modeled data as live.
Local time
—
—
Data connection
—
Last live scan
—
Next refresh
—
Scheduler runs while the local service is open.
Best market now
—
Multi-source live data mesh
Free/public discovery first, scarce paid/free-tier calls only for verification. Castle deduplicates records and labels exactly which source supports each lead.
Source health
Waiting for local service…
Quota guard
Waiting for local service…
Evidence rule
Free source finds it → second source cross-checks it → scarce API verifies only the best candidates. A single-source listing never becomes “high confidence” just because it looks attractive.
Live market ranking
Score is based on recent buyer-organization purchase proxy, DOM, price reductions, distressed listing share, recent transaction depth, and evidence size.
| Market | Score | Investor proxy | Median DOM | Listings cut ≥5% | Distress share | Evidence | Updated |
|---|
Today’s call queue — ranked by expected value
These are current listed opportunities with an actual agent/office contact when available. Off-market owner-direct contact needs a skip-trace/contact data source and is never invented.
| # | Property | Why it surfaced | Call | Ask | Open / Target / Walk | Buyer disposition | Profit | Model close | Evidence | Action |
|---|
Call Brief
SELECT A LEADSelect a ranked property. Castle will show who to call, what the evidence actually says, what to ask, opening/target/walk numbers, and what would make us pass.
Buyer Exit Hunter
WAITINGAfter selecting a lead, Castle scans recent purchases in the ZIP and ranks organization buyers by transaction evidence and price fit.
Live Autopilot requires an authorized property-data feed. The system is fully wired for RentCast live listings/property/AVM data; direct owner phone/email and buyer contact/POF still require contact-enrichment data. Castle will show “needs enrichment” rather than make them up.
Call Command Center
Cortex adapts to seller mood and intent while protecting consent, evidence and financial rails. Every completed call is organized into its property/session archive.
MOOD UNKNOWNSHADOW
Seller local time
—
—
Call timing
—
—
Telnyx
—
—
AI Coach
—
—
Today's ranked call queue
Lead quality × deal economics × evidence × timing. Calls outside the legal/local window are blocked.
Call control
Select a lead or enter a number. Castle requires traceable consent evidence for AI-generated calls and blocks unknown DNC status or prohibited local times.
Turn-taking rule: The seller hears a short automated hold disclosure while your required listen-only monitor leg rings. After you answer, Castle greets once, listens for a completed turn, and replies once. Retry Monitor redials only the monitoring leg; Take Over switches you into the conversation.
Deal Brain — memory, research and negotiation rails
Remembers this seller/property across calls, uses approved lessons from outcomes, and protects opening / target / approval / walk-away. Research is advisory and never overrides verified deal data.
Negotiation rails
—Prior calls
0Learning
0 outcomesResearch
Not runLearning rule: repeated post-call lessons become candidates only. They affect future calls only after you approve them; Castle never silently rewrites its own playbook.
Live Coach Brain
Only you see this. It protects the numbers, detects objections/closing signals, and tells you the next useful question—not pressure tactics.
Next move
Start or select a call.
Suggested line
—
Warning / guardrail
—
Seller priority
—
Objection
—
Closing signal
—
Instant Verify
Castle uses the first configured web-capable AI provider (Gemini first by default, then Groq/OpenAI if enabled). Property-specific facts should still prefer property/public-record APIs when available.
Transcript
No live transcript yet.
Post-call review
After hang-up, Castle scores discovery, listening, clarity, financial discipline, objection handling and the close.
Seller CRM + Smart Follow-up
One seller record across calls, email, SMS and future follow-ups. Castle pauses low-value leads instead of chasing forever.
Pursuit score
—Attempts
—Next action
—Status
—Castle will explain whether this lead still deserves attention.
Smart Email / SMS Follow-up
No-spam rule: Castle does not cold-blast email/SMS. Automated messages are for existing seller conversations or seller-requested follow-ups, and SMS requires recorded consent.
Seller engagement timeline
No seller activity recorded yet.
Timing learned from our calls
Follow-ups due
Calendar & Communications
One seller timeline across callbacks, email, SMS, contract deadlines and closing milestones. Castle schedules the next useful action and deliberately avoids duplicate or unnecessary contact.
Due today
0
Next 7 days
0
Queued messages
0
Overdue
0
Add calendar event
Smart communication composer
Uses the current seller record. Sending is blocked by suppression/consent rules; queued messages appear automatically on the calendar.
Contact Governor: no automatic cold blasting. SMS needs recorded consent; STOP/DNC suppresses future marketing contact; seller-requested timing overrides generic cadence.
Upcoming schedule
Manual events + seller callbacks + queued email/SMS + current transaction deadlines.
Loading schedule…
Communication queue
No queued messages.
Calendar rules
Seller asked for a time? That exact callback wins.
Contacted recently? Do not stack another channel unnecessarily.
No economic progress? Space attempts out or pause the lead.
Contract deadline? Transaction deadlines outrank nurture follow-ups.
Shadow Validation Lab
Short gate only: default 5–10 validation calls or maximum 2 days. Real autonomous AI calling and auto-send stay blocked until the guardrails survive the test set.
Status
—
Completed
0
Target 8
Pass score
0%
Critical failures
0
Calibration
LEARNING
Need real outcomes.
Mock Call Lab
These scenarios try to make Castle do something financially stupid or operationally annoying.
Run a scenario. Castle records whether the system protected the deal correctly.
Graduation gate
Shadow Mode is intentionally short. Passing the mock suite does not prove profitability; it proves the software is less likely to make obvious acquisition mistakes before real calls begin.
Live 10-Call Validation
Use consenting testers only. This bypasses Shadow Mode only for explicitly marked validation calls; real autonomous seller calls remain blocked.
Choose a property and seller behavior.
Live validation result
Completed
0/10Passed
0Call status
IDLETakeover
—Validation history
Critical scenarios include hidden repairs, title issues, co-owners, tenant occupancy, weak evidence and prices above walk-away.
| Scenario | Critical | Result | Expected behavior | Why | Time |
|---|
Deal Desk
All deep underwriting lives here. Autopilot finds the opportunity; Deal Desk proves whether the money survives reality.
Current property
—
Loaded from Autopilot or entered manually.
Opening offer
—
Walk-away
—
Base net
—
Current decision
Load a property from Autopilot or Property Brain.
Morning Command Center
The whole business condensed into one screen: best market, best lead, safest offer, buyer support, money at risk, and the single next action.
Best loaded market
—
Best lead
—
Safe walk-away
—
Do not cross without new evidence.
Base net profit
—
Decision confidence
—
Brutal decision
Run the engine.
Deal killers
Do these next
Evidence health
Activity
Market Radar
Ranks only the data you have loaded. Score rewards buyer activity, seller leverage and liquidity, and penalizes stale/weak evidence.
| Market | Investor share | Seller surplus | Price YoY | DOM | Inventory YoY | Score | Freshness | Evidence |
|---|
Scoring weights
Investor activity27%
Seller surplus23%
Price softness14%
Days on market13%
Inventory13%
Evidence/freshness10%
A market score is a hunting priority, not a guarantee that a seller will accept or a buyer will close.
Lead Engine
Ranks property-level motivation and economic fit without treating personal hardship as a sales weapon.
Lead score
—Model close
—Tier
—Action
—Ranked leads
Close model is deliberately capped and should be recalibrated from your own wins/losses. It is not a promise.
Property Brain
The subject-property truth layer. It combines manual facts with Comp Engine and Repair Estimator outputs, then measures evidence confidence.
ARV range
—
Repair range
—
Evidence confidence
—
Title risk
—
Hard gate until verified.
Comp Engine 2.0
Weighted ARV from comparable sales. Fresh, nearby, similar comps carry more weight. Outliers get less influence instead of silently distorting the deal.
| Use | Comp | Sale price | Sqft | Days ago | Miles | Similarity 1–5 | $/sqft | Weight | Implied subject |
|---|
Weighted ARV
—
Low / high
—
Evidence score
—
Selected comps
—
3+ strong comps preferred.
Comp notes
Use sold comps when possible. Similarity should reflect beds/baths, age, condition, lot, garage, layout and neighborhood—not just distance.
Repair Estimator
Builds low/base/high rehab totals from component-level assumptions instead of typing one convenient round number.
| Include | Category | Low | Base | High | Evidence | Note |
|---|
Low rehab
—
Base rehab
—
High rehab
—
Repair evidence
—
Photos/contractor evidence should raise confidence.
Offer Intelligence
The money page. It separates what the seller wants, what the deal can support, what the buyer can really pay, and what we make in downside/base/upside cases.
Offer ladder
Seller currently asks—
Opening offer—
Target contract—
Absolute walk-away—
Likely buyer close—
Best loaded buyer—
Downside
—
Base
—
Upside
—
Run the model.
Sensitivity matrix: net profit
Rows change ARV; columns change rehab. Red cells tell you how quickly the deal breaks.
Math transparency
Deal killers
Buyer Evidence
A buyer is not “real” because they say they buy. Match score combines ZIP fit, price fit, recent closes, proof of funds, reliability, retrade history and evidence freshness.
| Buyer | Strategy | ZIPs | Hard max | Recent closes | POF | Reliability | Retrade | Evidence date | Match |
|---|
Best exit
Buyer rules
Never choose purely by highest price. Expected value = realistic fee × close probability. A slightly lower buyer with clean POF and no retrades can be worth more.
Buyer Auction / Disposition Optimizer
Actual bids outrank modeled ceilings. Expected value penalizes retrades and weak reliability.
| Buyer | Price basis | Price | Gross fee | Reliability | Retrade | Expected net value | Rank |
|---|
Title & Ownership Gate
A motivated seller and good spread do not matter if ownership, payoff, liens, taxes or signing authority break the transaction.
Gate result
Not evaluated.
Why this matters
Castle treats unknown owner, unknown liens, unknown taxes/payoff and unverified signing authority as transaction risk. It cannot replace a title company or real-estate attorney.
Deal Room
One truth source for seller price, contract basis, buyer support, title, readiness, risks and final economics.
Seller ask
—
Contract basis
—
Likely buyer close
—
Base net
—
Model close score
—
Deal Kill Switch is waiting for analysis.
Readiness gates
Current deal brain
Closing Command Center
Contract to funded closing. Deadlines, title, deposits, buyer documents and access are visible in one place so a profitable deal does not die from admin.
Critical path
Closing readiness
0/8
Castle tracks the process; title/escrow/counsel remains the authoritative source for title and closing instructions.
Call Coach
Uses the actual offer ladder and seller context. It helps you diagnose the tradeoff; it does not manufacture desperation.
Verify every assumption in the conversation. If retail listing better fits the seller, say so.
Transaction & Disposition OS
From seller acceptance to title, disposition, secured buyer, closing and profit. Castle never treats an indication as a firm offer or an interested buyer as secured.
Stage
—
Buyer liquidity
—
Assignment probability
—
Model only · based on our evidence, never a guarantee.
Capital at risk
—
Best expected net
—
No exit quote yet.
Deal & contract facts
Record facts only. Castle does not invent legal language or mark a contract signed until you confirm it.
Pre-contract safety gate
Castle blocks “contract ready” if buyer demand is weak, economics exceed walk-away, legal/state pack is unconfirmed, title path is missing, payout readiness is unknown, or EMD is unavailable.
Disposition Desk
Direct offers, marketplace indications and JV quotes are kept separate. Only written/verified milestones raise close confidence.
| Use | Partner | Route | Quote type | Buyer / exit price | Our share % | Our net | POF | Signed | Buyer EMD | Close date | SLA / status | Actions |
|---|
No disposition evidence yet.
Contract-to-close center
Foreign individual / no-LLC readiness
An LLC is not a Castle requirement. If the transaction is legally allowed in your individual name, Castle can track it that way. But payout is never marked ready until the real title/closing company confirms your identity/KYC, signing method, tax documentation and the account it can legally pay.
Document Vault
Upload transaction documents into the local Castle folder. Keep attorney-approved legal templates separate from Castle-generated deal summaries.
Deadlines & SLA
Transaction activity
Pipeline
Tracks expected net, downside net, readiness and blockers. A large gross fee with a bad downside is not treated as a good deal.
| Deal | Stage | Ask | Contract | Base net | Downside net | Close model | Confidence | Primary blocker |
|---|
Finance Desk
Turns closed deals and expenses into unit economics: cost per lead, contract, close, average net fee, ROI and cash efficiency.
Closed deals
—
Gross assignment
—
Net profit
—
Operating spend
—
ROI
—
Expense ledger
Closed-Deal Learning
The system learns only from outcomes you actually record. It updates conversion statistics and flags which lead patterns and buyers deserve more trust.
Recorded outcomes
—
Win rate
—
Average closed net
—
Calibration strength
—
What our data currently says
Loss Analyzer
Every lost deal must teach the scoring engine something. This shows what actually kills our deals instead of blaming “bad leads.”
Documents & Closing Center
A control layer for the documents and evidence the transaction needs. It does not replace state-specific legal review.
Transaction checklist
Deal memo
Operator
Ask the workspace for a decision. Answers are generated from current data and formulas—not pretend internet research.
Operator
Try: “What should I offer?”, “How much can we sell it for?”, “Profit range”, “Should we walk?”, “Best buyer”, “Deal killers”, “Best market”, “What should I do next?”, “Are our comps strong?”, or “Repair risk”.
Try: “What should I offer?”, “How much can we sell it for?”, “Profit range”, “Should we walk?”, “Best buyer”, “Deal killers”, “Best market”, “What should I do next?”, “Are our comps strong?”, or “Repair risk”.
Fast commands
Castle Cortex
Control the talking intelligence, permanent memory, call archives, recording, maintenance and operator-approved improvement suggestions from one screen.
Cortex health
—
Not scanned
Permanent memories
0
Archived calls
0
Archive usage
0 GB
—
Organized Call Archive
Folders: year / month / day / property / session. Each call can contain transcript, memory, analysis, manifest and consented audio.
No calls archived yet.
Repair + Upgrade Suggestions
Cortex scans evidence and proposes changes. Approved means queued for operator review; production code is never silently rewritten.
Run a scan to generate suggestions.
System Integrity
Autosave, compliance, live integrations, evidence audit, backup/restore, and rules that prevent the UI from claiming more certainty than the inputs support.
Integration Center
Everything is centralized in CASTLE_CONFIG.json. You can edit it directly with EDIT_CASTLE_CONFIG.bat; secret keys never need to be pasted into chat. Test Connections can consume one real request from each configured external provider.
In AI mode this line is listen-only until you press Take Over.
Castle still discloses that the voice is automated; the name cannot be used to impersonate a human.
Editable in CASTLE_CONFIG.json.
Loading connection status…
API & Cost Governor
Tracks Castle-side requests, protected capacity, provider cooldowns, credential rotations and recent failures. Castle fingerprints keys one-way; it never displays the raw credential. A new API key does not automatically mean a fresh subscription quota.
Loading usage ledger…
RentCast quota checkpoint
RentCast quota belongs to the API subscription/billing period. If you rotate a key, Castle preserves the old quota usage until you confirm the current usage shown in your RentCast dashboard.
Use this only after checking the RentCast dashboard. Enter 0 only when the dashboard really shows a fresh quota.
Compliance profile
Integrity audit
Workspace snapshot
Operating rules
1. Current ask ≠ negotiated contract.
Profit uses actual contract only when one exists; otherwise it uses the current ask.
Profit uses actual contract only when one exists; otherwise it uses the current ask.
2. Upside never leads the decision.
Downside and evidence confidence are checked first.
Downside and evidence confidence are checked first.
3. Stale evidence loses weight.
Market snapshots, comps and buyers age out visibly.
Market snapshots, comps and buyers age out visibly.
4. Unknown title/compliance can block readiness.
No pretty score overrides a hard gate.
No pretty score overrides a hard gate.