How Propelligence works — in detail.

A property due-diligence tool. You tell it what matters to you — the commute, the room you need, the renovation you'd take on, your all-in ceiling — and it turns each listing you're weighing into a deep, defensible brief, scored against your own rubric. Its job is to help you not buy the wrong house.

Buyers weighing more than one property, and the agents helping them. If you're comparing a shortlist and want more than the listing tells you — zoning, comps, market history, what a renovation really costs — it's for you.

No. It never shows you homes to browse. You bring the listings you're already considering; Propelligence does the research the listing left out.

An AI assistant that speaks MCP (Model Context Protocol) — today that means Claude, via Claude Desktop, Claude Code, or Cowork. Propelligence connects to your assistant as an MCP server: it hands over your profile, your rubric, and the research skill, and your assistant does the work on your own machine.

Yes. Propelligence provides the rubric, the research skill, and the dashboard; your assistant provides the compute. That split is what keeps your data on your side of the line and the service inexpensive — the app itself runs no AI.

Two reasons. Privacy — your budget, your files, and your logged-in accounts never leave your computer. And quality — listing portals block anonymous fetching, so the only reliable way to capture a cover photo, the true listing status, and the real days-on-market is to browse on your own logged-in session.

A service that fetched listings from its own servers would produce thinner briefs with gaps in exactly those places. Your assistant, browsing as you, doesn't. That one capability — browsing your own session — is what decides brief quality.

The research skill is written for Claude today. The method is vendor-neutral — what to research and how to judge it is kept separate from the tool mechanics — so another MCP-capable assistant that can browse on your own session, write a real file, and run bounded parallel research could be supported. An engine without that first capability produces briefs with no cover photos and gaps in listing status; that's a documented consequence, not a surprise.

The link is added to your queue. Your connected assistant picks it up, pulls your profile and rubric over MCP, and researches the property against your rubric — comps, zoning, market history, condition, and the questions the listing left open. It captures a cover photo from your session, drafts any zoning-office letters, and writes the finished brief back over MCP. It lands on your dashboard, scored and ranked against everything else you're weighing.

No — your assistant does. The hosted app stores your rubric and the returned briefs, and runs the scoring and ranking math in your browser. It does no research and no model compute of its own; that's precisely why it can be private and inexpensive.

When your plans depend on something the listing can't answer — an addition, an ADU, a short-term rental — your assistant drafts the letter to the town's zoning office so you can ask before you offer, instead of assuming. What you can build is a separate question from what's listed.

The mobile app can't run MCP, so a link you add on mobile is queued and runs when your desktop reconnects. A lighter brief can be produced on-device in the meantime and re-runs in full once your desktop is back — you're never blocked, and you never lose the request.

You're notified in-app and by email the moment a brief finishes and loads back into your dashboard.

Yes. Each generated brief is a single self-contained file that opens with no network. The only thing that ever reaches the internet is the optional map, and only after you press the button — so a brief can be emailed and opened anywhere.

The first run just asks you to describe what you're looking for, the way you'd tell a real-estate agent — where, what kind of place, what it's for, what you'd never want. It sorts that into a profile, reflects it back so you can correct it, and asks only about whatever you left out: purpose (a home, a weekend place, a rental), the community you want, the activities you care about, travel, pets, dealbreakers. From that it builds your rubric — the dimensions that matter to you and a starting weight for each — which you tune in the dashboard. You answer once — and it keeps learning: as you talk through properties it offers to save new preferences it picks up (never ones you didn't say), so your rubric sharpens over time instead of going stale.

Each property gets a 0–10 score — a weighted blend across the dimensions that matter to you: location, condition, value, budget fit, plus whatever else you named — a commute, schools, water, climate risk, room for pets. Your weights are a fixed pool that always totals 100%, so raising one lowers the others and every trade-off is visible. Scores always carry one decimal.

That's the point. Move any weight and the whole shortlist re-ranks live. Lock a dimension to hold it while you adjust the rest. The ranking answers "best property under my priorities," not the listing's.

Set your all-in ceiling and budget fit is scored against it. Describe the renovation you'd actually take on and it returns a cost estimate, folds it into the all-in price, and re-scores budget fit. Some things money can't fix — a waterfront you can't use, a program the lot can't reach — are capped, and always labelled with the reason, so a score never quietly disagrees with itself.

A score isn't a verdict; it's a running position. As you learn more — a disclosure, an inspection, a walk-through, an answer from the town — the analysis takes it in and the score re-measures, and you can see exactly what moved the reading. Two things sharpen this way — each brief as facts land, and your profile as you learn what you want — so the analysis compounds the more you use it.

Yes. Once a property is on your dashboard, Propelligence monitors the live listing for you — a price cut, a change in status (pending, back on market, sold), a relist that resets the public clock. Each event is folded into the brief and re-scored against your rubric, so the read reflects where the property stands today, not the day you added it.

When you're ready to move, Propelligence turns the research into a position to hand your agent: where to open, what the comps support, where market history gives you leverage, and what the property could earn as a rental. It's agent-facing — and your ceiling never appears in it. The position is built from comps and market history, not from your preapproval.

Yes — send the whole dashboard or a single brief to a partner or your agent. Your budget and headroom stay yours; nothing seller-facing carries them.

The findings the listing leaves out — a days-on-market counter that restarts on a relist (61 days, not 12), a lot zoned for what's there now and not the addition you're picturing, and every "Unknown" on the seller's disclosure treated as a risk that becomes yours at closing.

Every finding is written the same way: a claim, the plain evidence behind it, and the one person who confirms it.

No. Your ceiling, preapproval, and headroom are runtime inputs — you enter them in your browser, they drive the scoring, and they're never written to the server. They can travel inside a share link if you choose to send one, but Propelligence doesn't keep them.

On your machine. The service holds your rubric and the finished briefs you choose to load back; the research itself runs in your assistant, against your own accounts, on your machine — not on our servers.

Only the single cover photo captured at research time — and never third-party listing photography there's no licence for.

Propelligence is in private beta; pricing is announced at launch. Because you bring your own AI and the service does no compute of its own, it's designed to stay inexpensive.

Any — there's no fixed property type and no preset checklist. If you can find a listing for it, you can point Propelligence at it.

Request beta access. We onboard in small groups so every brief gets read — and so we can help you set up the MCP connection to your assistant.