Grounded in real data
Prompts are enriched with structured inputs from county records, REAPI, Google Maps, and your own CRM before the model ever runs.
Four principles
How we implement AI, in four rules.
Human in the loop
Every AI output is a draft. Descriptions, scripts, and emails route through your agents for review before anything ships to a client.
Private by default
Team data is scoped by Row-Level Security. Your leads, deals, and buyers never train a foundation model and never cross team lines.
Auditable + logged
Every model run is written to an internal audit table with provider, endpoint, tokens, and duration. Nothing runs in the shadows.
Per-tool breakdown
What AI actually does — and where the data comes from.
Lead Evaluator
Deal analysis, cap-rate + DSCR reasoning, written investment memos
Model: Google Gemini 2.5 ProSource: REAPI · County records · Your inputs
Lead Generator
Market picks, owner outreach scripts, targeting rationale
Model: Google Gemini 2.5 FlashSource: REAPI · Public deed data
Skip Tracing Import
Owner match scoring, contact normalization, dedup suggestions
Model: Google Gemini 2.5 FlashSource: REAPI enrichment · USPS-verified addresses
Buyers Dashboard — copy
AI-written property descriptions in the Saratoga style
Model: Google Gemini 2.5 FlashSource: Your buy-box + property inputs
Buyers Dashboard — hero
Landing-page hero image generation from a prompt
Model: Gemini 2.5 Flash Image (Nano Banana)Source: Your prompt only
CRM
Draft outreach emails for cases and deals
Model: Google Gemini 2.5 FlashSource: Case/deal record + your saved signature
Quick Add — sign OCR
Read a yard-sign or house number from a field photo
Model: Google Gemini 2.5 Flash (vision)Source: Your photo + optional GPS
Quick Add — voice
Transcribe a spoken lead note to text
Model: OpenAI GPT-4o-mini-transcribeSource: Your voice recording only
Credible sources
Every answer traces back to a source you'd cite in a broker opinion.
AI never fabricates ownership, comparable sales, or contact info. When a fact isn't in the record, the model is instructed to say so — not to guess.
County assessor & recorder records
Ownership, parcel data, deeds, mortgages, and liens sourced via REAPI's licensed pipeline from public records.
Google Maps Geocoding & Places
Address verification and reverse-geocoding for the Quick Add sign-OCR flow and every mapped lead.
REAPI (Real Estate API)
Property attributes, owner records, mailing addresses, and skip-trace-quality contact enrichment.
Your first-party CRM data
Leads, deals, and buyers you and your team created. Scoped to your team by RLS. Never shared, never sold.
The stack
Models & APIs we recommend — and use in production.
All AI calls route through a single gateway with per-run logging, rate limiting, and a fixed model allowlist. Swapping a model is a governed change, not a silent upgrade.
Reasoning
Google Gemini 2.5 Pro
Long-context deal analysis and investment memos in Lead Evaluator.
Fast text
Google Gemini 2.5 Flash
Scripts, emails, property copy, OCR, and match scoring across the platform.
Images
Gemini 2.5 Flash Image (Nano Banana)
Hero images for Buyers Dashboard buy-box landing pages.
Speech
OpenAI GPT-4o-mini-transcribe
Voice-note transcription in Quick Add.
Geocoding
Google Maps Platform
Address verification, reverse-geocoding, and place lookups.
Property data
REAPI
Licensed access to county records, deeds, ownership, and skip-trace fields.
Guardrails & privacy
What we refuse to do with your data.
- Team data is scoped by Row-Level Security — never crosses team boundaries.
- No customer data is used to train foundation models.
- Every AI run is logged to a private audit table (provider, endpoint, tokens, latency).
- Prompts are versioned and reviewable; changes are governed, not silent.
- Model outputs are drafts — an agent always ships the final version.
- Credentials live server-side only; keys never ship to the browser.
Want to see the AI tools in action?
Walk through Lead Evaluator, Buyers Dashboard, and Quick Add on a live account — with the audit log open the entire time.